Christodoulou D. - Mathematical - Problems - of

SS
E
E
M
SS
E
E
M MM
zlamn_Christodoulou_titelei 3.1.2008 16:07 Uhr Seite 1
Zurich Lectures in Advanced Mathematics
Edited by
Erwin Bolthausen (Managing Editor), Freddy Delbaen, Thomas Kappeler (Managing Editor), Christoph Schwab,
Michael Struwe, Gisbert Wstholz
Mathematics in Zurich has a long and distinguished tradition, in which the writing of lecture notes volumes
and research monographs play a prominent part. The Zurich Lectures in Advanced Mathematics series
aims to make some of these publications better known to a wider audience. The series has three main con-
stituents: lecture notes on advanced topics given by internationally renowned experts, graduate text books
designed for the joint graduate program in Mathematics of the ETH and the University of Zurich, as well
as contributions from researchers in residence at the mathematics research institute, FIM-ETH. Moderately
priced, concise and lively in style, the volumes of this series will appeal to researchers and students alike,
who seek an informed introduction to important areas of current research.
Previously published in this series:
Yakov B. Pesin, Lectures on partial hyperbolicity and stable ergodicity
Sun-Yung Alice Chang, Non-linear Elliptic Equations in Conformal Geometry
Sergei B. Kuksin, Randomly forced nonlinear PDEs and statistical hydrodynamics in 2 space dimensions
Pavel Etingof, Calogero-Moser systems and representation theory
Guus Balkema and Paul Embrechts, High Risk Scenarios and Extremes - A geometric approach
Published with the support of the Huber-Kudlich-Stiftung, Zrich
SS
E
E
M
SS
E
E
M MM
European Mathematical Society
Demetrios Christodoulou
Mathematical
Problems of General
Relativity I
Author:
Prof. Demetrios Christodoulou
Department of Mathematics
ETH-Zentrum
8092 Zrich
Switzerland
2000 Mathematics Subject Classification (primary; secondary): 83C05; 35L70, 35Q75, 53C50, 58J45,
83C40, 83C45
ISBN 978-3-03719-005-0
The Swiss National Library lists this publication in The Swiss Book, the Swiss national bibliography,
and the detailed bibliographic data are available on the Internet at http://www.helveticat.ch.
This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, re-use of illustrations, recitation, broadcasting,
reproduction on microfilms or in other ways, and storage in data banks. For any kind of use permission
of the copyright owner must be obtained.
2008 European Mathematical Society

Contact address:
European Mathematical Society Publishing House
Seminar for Applied Mathematics
ETH-Zentrum FLI C4
CH-8092 Zrich
Switzerland
Phone: +41 (0)44 632 34 36
Email: info@ems-ph.org
Homepage: www.ems-ph.org
Cover picture: Black hole-powered jet of electrons and sub-atomic particles streams from center of galaxy
M87. Courtesy NASA and The Hubble Heritage Team (STScI/AURA)
Typeset using the authors T
E
X files: I. Zimmermann, Freiburg
Printed on acid-free paper produced from chlorine-free pulp. TCF
Printed in Germany
9 8 7 6 5 4 3 2 1
Contents
General introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 The laws of General Relativity . . . . . . . . . . . . . . . . . . . . . . . 9
2.1 The Einstein equations . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.1 Regular ellipticity and regular hyperbolicity . . . . . . . . . . 16
2.2 The Cauchy problem . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2.1 Cauchy problem for the Einstein equations: local in time,
existence and uniqueness of solutions . . . . . . . . . . . . . 19
2.3 Decomposition of the Einstein equations . . . . . . . . . . . . . . . 31
3 Asymptotic atness at spacelike innity and conserved quantities . . . . . 35
3.1 Conserved quantities . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2 Asymptotic atness . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.2.1 The maximal time function . . . . . . . . . . . . . . . . . . . 58
3.2.2 Positivity of the energy . . . . . . . . . . . . . . . . . . . . . 60
3.3 Angular momentum . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.3.1 Independence from exhaustion . . . . . . . . . . . . . . . . . 71
3.3.2 Conservation of angular momentum . . . . . . . . . . . . . . 71
3.4 The center of mass integrals . . . . . . . . . . . . . . . . . . . . . . 72
3.4.1 Conservation of center of mass integrals . . . . . . . . . . . . 76
4 The global stability of Minkowski spacetime . . . . . . . . . . . . . . . . 78
4.1 Statement of the problem . . . . . . . . . . . . . . . . . . . . . . . 78
4.1.1 Field theories in a given spacetime . . . . . . . . . . . . . . . 79
4.2 Sketch of the proof of the global stability of Minkowski spacetime . 103
4.2.1 The problem in General Relativity two main difculties . . . 103
4.2.2 Resolution of the rst difculty . . . . . . . . . . . . . . . . . 103
4.2.3 Resolution of the second difculty . . . . . . . . . . . . . . . 109
4.2.4 The controlling quantity . . . . . . . . . . . . . . . . . . . . 124
4.2.5 The continuity argument . . . . . . . . . . . . . . . . . . . . 127
4.2.6 Estimates for the geometric quantities associatedtothe maximal
foliation {H
t
] . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.2.7 Estimates for the geometric quantities associated to the null
foliation {C
u
] . . . . . . . . . . . . . . . . . . . . . . . . . . 132
4.2.8 Decomposition of a Weyl eld with respect to the surfaces S
t,u
137
4.2.9 The borderline error integrals . . . . . . . . . . . . . . . . . . 139
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
General introduction
General Relativity is a theory proposed by Einstein in 1915 as a unied theory of
space, time and gravitation. The theorys roots extend over almost the entire previous
history of physics and mathematics.
Its immediate predecessor, Special Relativity, established in its nal form by
Minkowski in 1908, accomplished the unication of space and time in the geometry
of a 4-dimensional afne manifold, a geometry of simplicity and perfection on par
with that of the Euclidean geometry of space. The root of Special Relativity is
Electromagnetic Theory, in particular Maxwells incorporation of Optics, the theory
of light, into Electrodynamics.
General Relativity is based on and extends Newtons theory of Gravitation as
well as Newtons equations of motion. It is thus fundamentally rooted in Classical
Mechanics.
Perhaps the most fundamental aspect of General Relativity however, is its geo-
metric nature. The theory can be seen as a development of Riemannian geometry,
itself an extension of Gauss intrinsic theory of curved surfaces in Euclidean space.
The connection between gravitation and Riemannian geometry arose in Einsteins
mind in his effort to uncover the meaning of what in Newtonian theory is the fortuitous
equality of the inertial and the gravitational mass. Identication, via the equivalence
principle, of the gravitational tidal force with spacetime curvature at once gave a
physical interpretation of curvature of the spacetime manifold and also revealed the
geometrical meaning of gravitation.
One sees here that descent to a deeper level of understanding of physical reality is
connected with ascent to a higher level of mathematics. General Relativity constitutes
a triumph of the geometric approach to physical science.
But there is more to General Relativity than merely a physical interpretation of
a variant of Riemannian Geometry. For, the theory contains physical laws in the
form of equations Einsteins equations imposed on the geometric structure. This
gives a tightness which makes the resulting mathematical structure one of surpassing
subtlety and beauty. An analogous situation is found by comparing the theory of
differentiable functions of tworeal variables withthe theoryof differentiable functions
of one complex variable. The latter gains, by the imposition of the CauchyRiemann
equations, a tighter structure which leads to a greater richness of results.
The domain of application of General Relativity, beyond that of Newtonian the-
ory, is astronomical systems, stellar or galactic, where the gravitational eld is so
strong that it implies the potential presence of velocities which are not negligible in
comparison with the velocity of light. The ultimate domain of application is the study
of the structure and evolution of the universe as a whole.
viii General introduction
General Relativity has perhaps the most satisfying structure of all physical the-
ories from the mathematical point of view. It is a wonderful research eld for a
mathematician. Here, results obtained by purely mathematical means have direct
physical consequences.
One example of this is the incompleteness theoremof R. Penrose and its extensions
due to Hawking and Penrose known as the singularity theorems". This result is
relevant to the study of the phenomenon of gravitational collapse. It shall be covered
in the second volume of the present work. The methods used to establish the result
are purely geometrical the theory of conjugate points. In fact, part of the main
argument is already present in the theory of focal points in the Euclidean framework,
a theory developed in antiquity.
Another example is the positive energy theorem, the rst proof of which, due to
R. Schoen and S. T. Yau, is based on the theory of minimal surfaces and is covered
in the the present volume. In this example a combination of geometric and analytic
methods are employed.
A last example is the theory of gravitational radiation, a main theme for both
volumes of this work. Here also we have a combination of geometric and analytic
methods. A particular result in the theory of gravitational radiation is the so-called
memory effect [11], which is due to the non-linear character of the asymptotic laws at
future null innity and has direct bearing on experiments planned for the near future.
This result will also be covered in our second volume.
The laws of General Relativity, Einsteins equations, constitute, when written in
any system of local coordinates, a non-linear system of partial differential equations
for the metric components. Because of the compatibility conditions of the metric with
the underlying manifold, when piecing together local solutions to obtain the global
picture, it is the geometric manifold, namely the pair consisting of the manifold itself
together with its metric, which is the real unknown in General Relativity.
The Einstein equations are of hyperbolic character, as is explained in detail in this
rst volume. As a consequence, the initial value problem is the natural mathematical
problem for these equations. This conclusion, reached mathematically, agrees with
what one expects physically. For, the initial value problem is the problem of deter-
mining the evolution of a system from given initial conditions, as in the prototype
example of Newtons equations of motion. The initial conditions for Einsteins equa-
tions, the analogues of initial position and velocity of Newtonian mechanics, are the
intrinsic geometry of the initial spacelike hypersurface and its rate of change under a
virtual normal displacement, the second fundamental form. In contrast to the case of
Newtonian mechanics however, these initial conditions are, by virtue of the Einstein
equations themselves, subject to constraints, and it is part of the initial value problem
in General Relativity a preliminary part to analyze these constraints. Important
results can be obtained on the basis of this analysis alone and the positive energy
theorem is an example of such a result.
General introduction ix
An important notion in physics is that of an isolated system. In the context of the
theory of gravitation, examples of such systems are a planet with its moons, a star
with its planetary system, a binary or multiple star, a cluster of stars, a galaxy, a pair or
multiplet of interacting galaxies, or, as an extreme example, a cluster of galaxies but
not the universe as a whole. What is common in these examples is that each of these
systems can be thought of as having an asymptotic region in which conditions are
trivial. Within General Relativity the trivial case is the at Minkowski spacetime of
Special Relativity. Thus the desire to describe isolated gravitating systems in General
Relativity leads us to consider spacetimes with asymptotically Minkowskian regions.
However it is important to remember at this point the point of view of the initial
value problem: a spacetime is determined as a solution of the Einstein equations
from its initial data. Consequently, we are not free to impose our own requirements
on a spacetime. We are only free to impose requirements on the initial data to the
extent that the requirements are consistent with the constraint equations. Thus the
correct notion of an isolated systemin the context of General Relativity is a spacetime
arising fromasymptotically at initial conditions, namely an intrinsic geometry which
is asymptotically Euclidean and is a second fundamental form which tends to zero at
innity in an appropriate way. This is discussed in detail in this volume.
Trivial initial data for the Einstein equations consists of Euclidean intrinsic ge-
ometry and a vanishing second fundamental form. Trivial initial data gives rise to the
trivial solution, namely the Minkowski spacetime. A natural question in the context
of the initial value problem for the vacuum Einstein equations is whether or not every
asymptotically at initial data which is globally close to the trivial data gives rise
to a solution which is a complete spacetime tending to the Minkowski spacetime at
innity along any geodesic. This question was answered in the afrmative in the joint
work of the present author with Sergiu Klainerman, which appeared in the monograph
[14]. One of the aims of the present work is to present the methods which went into
that work in a more general context, so that the reader may more fully understand
their origin and development as well as be able to apply them to other problems. In
fact, problems coming fromelds other than General Relativity are also treated in the
present work. These elds are Continuum Mechanics, Electrodynamics of Continu-
ous Media and Classical Gauge Theories (such as arise in the mesoscopic description
of superuidity and superconductivity). What is common to all these problems from
our perspective is the mathematical methods involved.
One of the main mathematical methods analyzed and exploited in the present
work is the general method of constructing a set of quantities whose growth can be
controlled in terms of the quantities themselves. This method is an extension of the
celebrated theorem of Noether, a theorem in the framework of the action principle,
which associates a conserved quantity to each 1-parameter group of symmetries of
the action (see [12]). This extension is involved at a most elementary level in the
very denition of the notion of hyperbolicity for an EulerLagrange system of partial
x General introduction
differential equations, as discussed in detail in this rst volume. In fact we may
say that such a system is hyperbolic at a particular background solution if linear
perturbations about this solution possess positive energy in the high frequency limit.
The application of Noethers Principle to General Relativity requires the intro-
duction of a background vacuum solution possessing a non-trivial isometry group,
as is explained in this rst volume. Taking Minkowski spacetime as the background,
we have the symmetries of time translations, space translations, rotations and boosts,
which give rise to the conservation laws of energy, linear momentum, angular momen-
tum and center of mass integrals, respectively. However, as is explained in this rst
volume, these quantities have geometric signicance only for spacetimes which are
asymptotic at innity to the background Minkowski spacetime, so that the symmetries
are in fact asymptotic symmetries of the actual spacetime.
The other main mathematical method analyzed and exploited in the present work
is the systematic use of characteristic (null) hypersurfaces. The geometry of null hy-
persurfaces has already been employed by R. Penrose in his incompleteness theorem
mentioned above. What is involved in that theorem is the study of a neighborhood
of a given null geodesic generator of such a hypersurface. On the other hand, in the
work on the stability of Minkowski spacetime, the global geometry of a characteristic
hypersurface comes into play. In addition, the properties of a foliation of spacetime
by such hypersurfaces, also come into play. This method is used in conjunction with
the rst method, for, such characteristic foliations are used to dene the actions of
groups in spacetime which may be called quasi-conformal isometries, as they are
globally as close as possible to conformal isometries and tend as rapidly as possible
to conformal isometries at innity. The method is introduced in this rst volume
and will be treated much more fully in the second volume. It has applications be-
yond General Relativity to problems in Fluid Mechanics and, more generally, to the
Mechanics and Electrodynamics of Continuous Media.
This book is based on Nachdiplom Lectures held at the Eidgenssische Tech-
nische Hochschule Zurich during the Winter Semester 2002/2003. The author wishes
to thank his former student Lydia Bieri for taking the notes of this lecture, fromwhich
a rst draft was written, and for making the illustrations.
1 Introduction
The general theory of relativity is a unied theory of space, time and gravitation. The
fundamental concept of the theory is the concept of a spacetime manifold.
Denition 1. A spacetime manifold is a 4-dimensional oriented differentiable mani-
fold M, endowed with a Lorentzian metric g.
Denition 2. A Lorentzian metric g is a continuous assignment of a non-degenerate
quadratic form g
]
, of index 1, in T
]
M at each M.
Here we denote by T
]
M the tangent space to M at . Also, non-degenerate
means
g
]
(X. Y ) = 0 V Y T
]
M == X = 0.
while of index 1 means that the maximal dimension of a subspace of T
]
M, on which
g
]
is negative denite, is 1.
An equivalent denition is the following.
Denition 3. Aquadratic formg
]
in T
]
M is called Lorentzian if there exists a vector
V T
]
M such that g
]
(V. V ) < 0 while setting
V
= {X : g
]
(X. V ) = 0] (the g
]
-orthogonal complement of V ),
g
]
[
1
is positive denite.
We can then choose an orthonormal frame (1
0
. 1
1
. 1
2
. 1
3
) at , by setting
1
0
=
V
p
g(V. V )
and choosing an orthonormal basis (1
1
. 1
2
. 1
3
) for
V
. Given any vector X T
]
M
we can expand
X = X
0
1
0
X
1
1
1
X
2
1
2
X
3
1
3
=
X
(j = 0. 1. 2. 3).
Then
g(1
. 1
) = j
= diag(1. 1. 1. 1).
g(X. X) = (X
0
)
2
(X
1
)
2
(X
2
)
2
(X
3
)
2
=
X
.
2 1 Introduction
Denition 4. The null cone at M,
N
]
= {X = 0 T
]
M : g
]
(X. X) = 0].
is a double cone N
]
= N
]
L N
-
]
.
Denote by 1
]
the interior of N
]
and by 1
-
]
the interior of N
-
]
.
Denition 5. The set of timelike vectors at M is dened as
1
]
:= 1
]
L 1
-
]
= {X T
]
M : g
]
(X. X) < 0].
where 1
]
is an open set.
Denition 6. The set of spacelike vectors at M is dened as
S
]
:= {X T
]
M : g
]
(X. X) > 0].
where S
]
is the exterior of N
]
, a connected open set.
Time orientability. We assume that a continuous choice of positive (future) compo-
nent 1
]
of 1
]
at each M, is possible. Once such a choice has been made, the
spacetime manifold M is called time oriented.
Denition 7. A causal curve in M is a differentiable curve ; whose tangent vector
; at each point M belongs to 1
]
L N
]
, that is, it is either timelike or null.
Then either ;
]
1
]
L N
]
at each along ; in which case ; is called future-
directed, or ;
]
1
-
]
LN
-
]
at each along ; in which case ; is called past-directed.
Given a point M, we can then dene the causal future of .
Denition 8. The causal future of , denoted by J
(), is the set of all points q M

for which there exists a future-directed causal curve initiating at and ending at q.
Similarly, we can dene J
-
(), the causal past of .
Denition 9. The arc length of a causal curve ; between the points corresponding
to the parameter values z = a. z = b is
1;|(a. b) =
Z
b
o
p
g( ;(z). ;(z)) Jz.
If q J
(), we dene the temporal distance of q from by

t(q. ) = sup
all future-directed causal
curves from ) to q
1;|.
The arc length is independent of the parametrization of the curve.
1 Introduction 3
Let us recall the HopfRinow theorem in Riemannian geometry:
Theorem 1 (HopfRinow). On a complete Riemannian manifold any two points can
be joined by a minimizing geodesic.
In Lorentzian geometry the analogous statement is in general false. It holds how-
ever when the spacetime admits a Cauchy hypersurface (the denition of this concept
will follow). When the supremum in the above denition is achieved and the metric
is C
1
the maximizing curve is a causal geodesic; after suitable reparametrization the
tangent vector is parallelly transported along the curve.
Examples of spacetime manifolds. We take as our model the
Riemannian spaces of constant curvature:
1
(1
k
4
[.[
2
)
2
[J.[
2
. k = 0. 1. 1.
where [ [ is the Euclidean magnitude, [[ =
p
P
i
(
i
)
2
. When k = 1 the
manifold is the ball of radius 2 in R
n
, [.[ < 2.
By analogy, we have the
Lorentzian spaces of constant curvature:
1
(1
k
4
(.. .))
2
(J.. J.).
Here ( . ) is the Minkowski quadratic form. (u. ) = u
0
P
n-1
i=1
u
i
i
.
For k = 1 we have what is called de-Sitter-space while for k = 1 we
have what is called Anti-de-Sitter-space. In the case k = 1 the manifold is
{. R
n
: (.. .) < 4].
.
0
. = (.
1
. . . . . .
n-1
)
o
o
q
Figure 1
4 1 Introduction
InAnti-de-Sitter space there are points and q as shown for which t(. q) = o.
For, the length of the timelike segment of the causal curve joining and q in the
gure can be made arbitrarily large by making the segment approach the hyperboloid
at innity (.. .) = 4.
Denition 10. We dene the causal future J
(1) and the causal past J

-
(1) of any
set 1 M, in particular a closed set, by
J
(1) = {q M : q J
() for some 1].

The boundaries of J
(1), J
-
(1), i.e. dJ
(1), dJ
-
(1) for closed sets 1,
are null hypersurfaces. They are realized as level sets of functions u satisfying
the eikonal equation g
ud
u = 0. These hypersurfaces are generated by null

geodesic segments, as shall be shown below. They are thus analogous to ruled surfaces
in Euclidean geometry. Moreover, the null geodesics generating J
(1) have past

end-points only on 1 and those generating J
-
(1) have future end-points only on 1.
The null geodesics generating J
(1) may have future end-points, even when 1 is a

single point. The set of these end-points forms the future null cut locus corresponding
to 1. Similarly for J
-
(1). (Null cut loci shall be discussed at length in the second
volume.)
Figure 2. The future null cut locus of a point . Each of the points marked by a dot is a point
where a pair of null geodesics issuing from intersect.
Denition 11. H is called a null hypersurface if at each point . H the induced
metric g
x
T
\!
is degenerate.
Thus there exists an 1 = 0 T
x
H such that
g
x
(1. X) = 0 V X T
x
H.
1 Introduction 5
Now T
x
H is a hyperplane in T
x
M. Such a hyperplane is dened by a covector
T
+
x
M,
T
x
H = {X T
x
M : X = 0].
Representing H as the (0-)level set of a function u, we can take
= Ju(.).
If we set 1
= g
u (components with respect to an arbitrary frame), we have

g(1. X) = Ju X
and 1 is g-orthogonal to H. Then H is a null hypersurface if and only if
1
x
T
x
H V . H.
Now let u be a function, each of the level sets of which is a null hypersurface.
Taking then X = 1 we obtain
g(1. 1) = 0 (g
= 0).
hence 1 is at each point a null vector, a condition which, expressed in terms of Ju,
reads
g
ud
u = 0.
which is the eikonal equation.
In fact, 1 is a geodesic vector eld, that is, the integral curves of 1 are null
geodesics. The proof of this fact is as follows.
(V
1
1)
= 1
. (1)
g
2
(V
1
1)
= 1
1
2
. (2)
where 1
2
= g
2
1
= d
2
u. Now the Hessian of a function is symmetric:
V
(d
2
u) = V
2
(d
u). (3)
Thus,
1
1
2
= 1
V
2
1
= d
2
1
2
g(1. 1)
= 0.
(4)
that is,
V
1
1 = 0. (5)
6 1 Introduction
Denition 12. A hypersurface H is called spacelike if at each . H, the induced
metric
g
x
T
\
1
=: g
x
is positive denite.
Then (H. g) is a proper Riemannian manifold. The g-orthogonal complement of
T
x
H is a 1-dimensional subspace of T
x
M on which g
x
is negative denite. Thus
there exists a vector N
x
1
x
of unit magnitude
g
x
(N
x
. N
x
) = 1
whose span is this 1-dimensional subspace. We call N (the so-dened vector eld
along H) the future-directed unit normal to H.
Denition 13. The 2
nd
fundamental form k of H is a 2-covariant symmetric tensor
eld on H, or quadratic form in T
x
H at each . H, dened by
k(X. Y ) = g(V
A
N. Y ) V X. Y T
x
H. (6)
Denition 14. A Cauchy hypersurface is a complete spacelike hypersurface H in M
(i.e., (H. g) is a complete Riemannian manifold) such that if ; is any causal curve
through any point M, then ; intersects H at exactly one point.
Examples of Cauchy hypersurfaces
A spacelike hyperplane in Minkowski spacetime M is a Cauchy hypersurface
for M.
A spacelike hyperboloid in Minkowski spacetime,
(.
0
)
2
3
X
i=1
(.
i
)
2
= 1. .
0
> 0.
is a complete Riemannian manifold (of constant negative curvature) but not a
Cauchy hypersurface for M. It is however a Cauchy hypersurface for 1
0
M.
The Anti-de-Sitter space does not admit a Cauchy hypersurface.
If we consider only future evolution the above denition is replaced by one in which
; is taken to be any past-directed causal curve. Then M is a manifold with boundary
and H is the past boundary of M.
Denition 15. A spacetime admitting a Cauchy hypersurface is called globally hy-
perbolic.
1 Introduction 7
Under the hypothesis of global hyperbolicity we can dene a time function. This
is a differentiable function t such that
Jt X > 0 VX 1
]
. V M. (7)
The manifold M is then diffeomorphic to the product R

M where

M is a 3-manifold,
each level set
t
of t being diffeomorphic to

M. H =
0
is a Cauchy hypersurface.
Denition 16. The lapse function corresponding to a time function t is the function
= (g
t d
t )
-
1
2
. (8)
This measures the normal separation of the leaves
t
(of the foliation by the level
sets of t ). Consider the vector eld
T
=
2
g
t. (9)
The integral curves of T are the orthogonal curves to the
t
-foliation. Moreover,
T t = T
t = 1. Thus the orthogonal curves are parametrized by t . That is, the 1-

parameter group
r
generated by T takes the leaves onto each other:
r
(
t
) =
tr
.
Thus T is a time translation vector eld. The unit normal N is given by
N =
-1
T. (10)
The integral curves of N are the same orthogonal curves but parametrized by arc
length s. It follows that, along an orthogonal curve,
Js
Jt
= . (11)
We can identify

M with
0
= H. The mapping of M into R

M or 0. o)

M,
taking M to the pair (t. q) if lies on
t
and along the orthogonal curve through
q
0
, is a diffeomorphism.
In terms of this representation of M we can write
g =
2
Jt
2
g. (12)
where g = g(t ) is the induced metric on
t
, which is positive denite. Moreover,
T =
d
dt
. N =
1
d
dt
.
Assume that (1
1
. 1
2
. 1
3
) is a frame eld for
t
, which is Lie transported along (the
integral curves of) T , that is,
T. 1
i
| = 0. i = 1. 2. 3. (13)
8 1 Introduction
(In particular we may take (.
1
. .
2
. .
3
) to be local coordinates on H =
0
, and set
1
i
=
J
Jx
i
.) Then we have
k
i}
= k(1
i
. 1
}
) =
1
2
(g(V
T
i
N. 1
}
) g(V
T
N. 1
i
))
=
1
2
(g(V
T
i
T. 1
}
) g(V
T
T. 1
i
))
=
1
2
(g(V
T
1
i
. 1
}
) g(V
T
1
}
. 1
i
))
=
1
2
T(g(1
i
. 1
}
)):
that is
k
i}
=
1
2
d g
i}
dt
. (14)
where g
i}
= g(1
i
. 1
}
) = g(1
i
. 1
}
) are the components of the induced metric on
t
. The above equation is called the 1
st
variational formula.
2 The laws of General Relativity
2.1 The Einstein equations
The laws of General Relativity are the Einstein equations linking the curvature of
spacetime to its matter content:
G
:= 1

1
2
g
1 = 2T
. (15)
(We are using rationalized units where 4 times Newtons gravitational constant as
well as the speed of light in vacuum are set equal to 1.) Here T
is the energy-
momentum tensor of matter, G
the Einstein tensor, 1
the Ricci tensor and 1 the

scalar curvature of the metric g
. From the original Bianchi identity

V
1
;je
:= V
1
;e
V
1
;e
V
;
1
e
= 0. (16)
one obtains
V
= 0. (17)
the twice contracted Bianchi identity. This identity (17) implies
V
= 0. (18)
the equations of motion of matter. The Einstein vacuum equations
G
= 0 (19)
correspond to the absence of matter: T
= 0. The equations are then equivalent to

1
= 0. (20)
The connection coefcients I
and the curvature and Ricci tensor components in

arbitrary local coordinates read as follows:
I
=
1
2
g
(d
). (21)
1
2
= d
2
I
2
I
2
I
2
. (22)
1
= 1
= d
. (23)
Denoting by P.P. the principal part, that is, the part containing the highest (2
nd
) deriva-
tives of the metric, we have
P.P.{1
] =
1
2
g
{d
]. (24)
10 2 The laws of General Relativity
We shall presently discuss the character of the Einstein equations as reected in their
symbol. The symbol is dened by replacing in the principal part
d
by
.
where
are the components of a covector and g
the components of a possible

variation of g. We then obtain the symbol o
at a point M and a covector

T
+
]
M, for a given background metric g:
(o
g)
=
1
2
g
).
Let us denote
(i
g)
= g
.
(. ) = g
.
( )
.
g
= tr g.
We can then write
(o
g) =
1
2
{ i
g i
g tr g (. ) g].
The notion of the symbol of a system of EulerLagrange equations is as follows. Let
us denote by ., the independent variables: .
, j = 1. . . . . n; by q, the dependent
variables: q
o
, a = 1. . . . . m; and by , the 1
st
derivatives of dependent variables:
o
,
n m matrices. Then the Lagrangian 1 is a given function of (.. q. ),
1 = 1(.. q. ).
A set of functions (u
o
(.) : a = 1. . . . . m) is a solution of the EulerLagrange
equations, if substituting
q
o
= u
o
(.).
=
du
o
d.
(.)
we have
d
d.
d1
d
o
(.. u(.). du(.))
d1
dq
o
(.. u(.). du(.)) = 0. (25)
Dening
o
=
d1
d
o
o
=
d1
dq
o
.
2.1 The Einstein equations 11
the EulerLagrange equations become
d
o
d.
=
o
.
The q, , , are analogous to position, velocity, momentumand force, respectively,
in classical mechanics.
The principal part of the EulerLagrange equations is
h
ob
d
2
u
b
d.
d.
(.. u(.). du(.)).

where
h
ob
=
d
2
1
d
o
d
b
(.. q. ).
Let us consider the equations of variation. These are the linearized equations, satised
by a variation through solutions. If we denote by u
o
the variations of the functions u
o
,
the principal part of the linearized equations is
h
ob
(.. u(.). du(.))
d
2
u
b
d.
d.
.
Consider in fact oscillatory solutions
u
o
= n
o
e
i
(26)
of the equations of variation. Writing

e
in place of , substituting in the linearized
equations and keeping only the leading terms as c 0 (high frequency limit), we
obtain
h
ob
(.. u(.). du(.)) n
b
d
d.
d
d.
= 0. (27)
The left-hand side is the symbol o
n, where
=
J
Jx
J
. Thus, the symbol of the
EulerLagrange equations is in general given by
(o
u)
o
= h
ob
u
b
=
ob
() u
b
. (28)
where
ob
() = h
ob
(29)
is an m m matrix whose entries are homogeneous quadratic polynomials in .
From a global perspective, the .
, j = 1. . . . . n are local coordinates on an

n-dimensional manifold M and . denotes an arbitrary point on M, while the q
o
,
a = 1. . . . . m are local coordinates on an m-dimensional manifold N and q denotes
an arbitrary point on N. The unknown u is then a mapping u: M N and the
functions (u
o
(.), a = 1. . . . . m) describe this mapping in terms of the given local
coordinates.
Denition 17. Let M be an n-dimensional manifold. Then the characteristic subset
C
+
x
T
+
x
M is dened by
C
+
x
= { = 0 T
+
x
M : null space of o
= 0]
= { = 0 T
+
x
M : det () = 0].
Thus C
+
x
if and only if = 0 and the null space of o
is non-trivial.
The simplest example of an EulerLagrange equation with a non-empty charac-
teristic is the linear wave equation
u := g
(d
u) = 0.
This equation arises from the Lagrangian
1 =
1
2
g
.
The symbol is o
u = (g
) u and the characteristic is

C
+
x
= { = 0 T
+
x
M : (. ) = g
= 0].
that is, C
+
x
is the null cone in T
+
x
M associated to the metric g.
Let us now return to the symbol for the Einstein equations. Let us set
g = (30)
for an arbitrary covector T
+
x
M. Then
i
g = (. )
=
J+
(
J
+
(. ) (31)
and
tr g = 2(. ). (32)
We see that
o
g = 0. (33)
Therefore the null space of o
is non-trivial for every covector . This degeneracy is

due to the fact that the equations are generally covariant. That is, if g is a solution
of the Einstein equations and is a diffeomorphism of the manifold onto itself, then
the pullback
+
g is also a solution. If X is a (complete) vector eld on M, then X
generates a 1-parameter group {
t
] of diffeomorphisms of M and
L
A
g =
J
Jt
+
t
g
t=0
. (34)
the Lie derivative with respect to X of g, is a solution of the linearized equations.
Let us recall that the Lie derivative of g with respect to a vector eld X is given
by
(L
A
g)(Y. 7) = g(V
Y
X. 7) g(V
7
X. Y ).
Setting Y = 1
and 7 = 1
, where (1
: j = 0. . . . . 3) is an arbitrary frame, we
can write
(L
A
g)
= V
.
where X
= g
2
X
2
. The symbol of a Lie derivative is given by
g
. where
=

X
.
A simple analogue. The Maxwell equations for the electromagnetic eld J
,
V
= g
2
V
2
J
= 0. (35)
provide a simple analogue to this situation. Let us recall that J = J, or J
=
d
, where
is the electromagnetic potential, a 1-form. The Maxwell

equations are the EulerLagrange equations of the Lagrangian
1 =
1
4
J
where J
= g
k
g
2
J
k2
. The symbol for these equations is
(o
= g
2
2
(
).
that is
o

= (.

) (. )

.
Consider the variation
= z
for any real number z. Then
o

= 0.
Thus we have a degeneracy here as well: the null space of o
is non-trivial for all

T
+
x
M. This is due to the gauge invariance of the Maxwell equations. If is a
solution, so is
= J
for any function . In fact

is considered to be equivalent to , just as
+
g is
considered to be equivalent to g. Thus (by linearity)

= d
is a solution of the
linearized equations, for any function . To remove the degeneracy we must factor
out these trivial solutions. Correspondingly in General Relativity we must factor out
the solutions of the formL
A
g = g for any vector eld X. At the level of the symbol
the gauge transformation is

.
So, we introduce the equivalence relation
1
~

2
=

1
z. z R.
We then obtain a quotient space Q of dimension 4 1 = 3. Consider now the
null space of o
with o
dened on Q. We distinguish two cases: in the rst case

(. ) = 0 and in the second case (. ) = 0.
Case 1. (. ) = 0. Then
o

= 0 ==

= z. z =
(.

)
(. )
.
that is

~ 0. Thus we have the trivial null space if is not a null covector.
Case 2. (. ) = 0. In this case we may choose another covector in the same com-
ponent of the null cone such that (. ) = 2. There is then a unique representative
in each equivalence class in Q such that

(.

) = 0.
For, take another element

t
out of the equivalence class of

, that is

t
=

z
for some z R. Then
0 = (.

t
) = (.

) 2z
implies that

is the unique representative of its equivalence class with (.

) = 0.
Let us work with this representation. Then it holds that
o

= (.

) = 0. = 0 (.

) = 0.
We conclude that the null space of o
consists of the spacelike 2-dimensional plane

, the g-orthogonal complement of the timelike plane spanned by and . So,
is the space of the degrees of freedom of the electromagnetic eld at a point (two
polarizations).
Returning to the symbol for the Einstein equations, the symbol for the Lie deriva-
tive
(L
A
g)
= V
reads, as noted above,
.
where the

X
are the components of an arbitrary covector. Consider then the equiv-

alence relation
g
1
~ g
2
g
2
= g
1
. T
+
x
M.
which gives a quotient space Q.
Again we distinguish the two cases according as to whether the covector satises
(. ) = 0 or (. ) = 0.
Case 1. (. ) = 0. If is not null, then o
g = 0 implies that
g = . where =
(i
g
1
2
tr g)
(. )
.
thus o
has only trivial null space on Q.

Case 2. (. ) = 0. If is null, we can choose in the same component of the null
cone N
+
x
in T
+
x
M such that (. ) = 2. There is then a unique representative g in
each equivalence class { g] Q such that
i
g = 0.
So,
o
g = 0 i
g i
g tr g = 0.
Taking the inner product with we see that (i
g. ) = (i
g. ) = 0, hence
2 i
g 2 tr g = 0.
Taking again the inner product with gives
4 tr g = 0. that is, tr g = 0.
Substituting this above yields
i
g = 0.
Conversely, i
g = i
g = 0 and tr g = 0 implies that g lies in the null space of o
.
Therefore, if N
+
x
, then the null space of o
can be identied with the space of

trace-free quadratic forms on the 2-dimensional spacelike plane , the g-orthogonal
complement of the linear span of and . This is the space of gravitational degrees
of freedom at a point (two polarizations).
2.1.1 Regular ellipticity and regular hyperbolicity. We have already introduced
the quadratic form
h
ob
=
d
2
1
d
o
d
b
(.. q. )
in the general context of a Lagrangian theory of mappings u: M N. A point
. M is represented in terms of local coordinates in M by (.
: j = 1. . . . . n).
The position q is a possible value of u(.), that is, a point in N, represented by
(q
o
: a = 1. . . . . m) in terms of local coordinates in N, while the velocity is a
possible value of Ju(.) and is represented by the n m-matrix
o
=
Ju
u
Jx
J
(.). Here
n = dimM and m = dimN.
Before stating the denition of regular ellipticity let us have a closer look at the
necessary notions. Let u: M N be a background solution, . M, q = u(.),
and let T
+
x
M, Q T
q
N. Then Q is a variation in position, the value at .
of a possible variation u of the background solution. The corresponding variation
in velocity is a linear map from T
x
M to T
q
N: L(T
x
M. T
q
N). For any
X T
x
M the components of the vector Q = X T
q
N are Q
o
=
o
, where
the X
are the components of X and the

o
are the components of . The space

S
2
(L(T
x
M. T
q
N)) of quadratic forms on L(T
x
M. T
q
N) splits into the direct sum
S
2
= S
2
S
2-
.
where S
2
consists of the even quadratic forms and S
2-
of the odd quadratic forms.
Thus, a quadratic form h on L(T
x
M. T
q
N) decomposes into
h = h
h
-
.
where h
andh
-
are, respectively, the evenandoddparts of h. Interms of components
we have
h
ob
= h
ob
h
-ob
.
where
h
bo
= h
ob
(h being a symmetric bilinear form), and
h
ob
= h
bo
= h
ob
.
h
-ob
= h
-bo
= h
-ob
.
We also need the following notion:
Denition 18. Rank-1-elements of L(T
x
M. T
q
N) are the elements of the form
= Q. T
+
x
M. Q T
q
N.
that is, X = ( X)Q for all X in T
x
M.
Now consider the quadratic form h( . ) = h
ob

o
on the velocity varia-

tions
o
.
Regular ellipticity (LegendreHadamard condition). A Lagrangian 1 is called
regularly elliptic at (.. q. ) if the quadratic formh =
J
2
1
J
2
(.. q. ) on L(T
x
M. T
q
N)
is positive denite on the set of rank-1-elements
o
Q
o
with T
+
x
M and
Q T
q
N.
If 1and 1
t
are two Lagrangians giving rise to the same EulerLagrange equations,
then the difference h h
t
of the corresponding quadratic forms is an odd quadratic
form.
Remark. The denition of regular ellipticity is independent of the choice of La-
grangian for the same EulerLagrange equations because odd quadratic forms vanish
on the set of rank-1-elements.
Next we dene regular hyperbolicity, a notion expounded in [12].
Denition 19. A Lagrangian 1 is called regularly hyperbolic at (.. q. ) if the
quadratic form h =
J
2
1
J
2
(.. q. ) on L(T
x
M. T
q
N) has the following property:
There exists a pair (. X) in T
+
x
M T
x
M with X > 0 such that:
1. h is negative denite on the space
1
= { Q : Q T
q
N].
2. h is positive denite on the set of rank-1-elements of the subspace
A
= { L(T
x
M. T
q
N) : X = 0].
Note that this denition is also independent of the choice of Lagrangian giving
rise to the same EulerLagrange equations.
Denition 20. Given a quadratic form h on L(T
x
M. T
q
N) and a pair (. X) in
T
+
x
MT
x
M withX >0, we dene a newquadratic formm(. X) onL(T
x
M.T
q
N)
depending linearly on and X by
m(. X)(
1
.
2
) = ( X)h(
1
.
2
) h(
1
X.
2
) h(
1
.
2
X). (36)
We call this the Noether transform of h dened by (. X).
Proposition 1. A Lagrangian 1is regularly hyperbolic at (.. q. ) if and only if there
exists a pair (. X) in T
+
x
M T
x
M with X > 0 such that the Noether transform
m(. X) of h corresponding to (. X) is positive denite on the set
1
= { 1 Q : V T
+
x
M. V 1. Q T
q
N]
(which is a set of special rank-2-elements).
Remark. If his an odd quadratic form, then the Noether transformof hcorresponding
to (. X) vanishes on 1
.
Given h and = 0 T
+
x
M, we dene (), a quadratic form in T
q
N, by
()(Q
1
. Q
2
) = h( Q
1
. Q
2
). (37)
That is,
ob
() = h
ob
.
Then the characteristic subset C
+
x
of T
+
x
M is dened by
C
+
x
= { = 0 T
+
x
M : () is singular ]. (38)
Also, given Q = 0 T
q
N, we dene a quadratic form (Q) in T
+
x
M as follows:
(Q)(
1
.
2
) = h(
1
Q.
2
Q). (39)
that is,
(Q) = h
ob
Q
o
Q
b
.
Next, for a given C
+
x
we dene
() = { (Q) : Q = 0 null space of ()]

= { X T
x
M : X = 0] (a hyperplane in T
x
M).
(40)
Here (Q) is considered as a linear map of T
+
x
M into T
x
M,
(Q)
.
() is a positive cone in
. That is, if X () and z > 0, then zX lies in ().

Also the following holds:
(j) = () V j > 0.
If is a regular point of C
+
x
, then the null space of () has dimension 1 and () is
a ray. Otherwise the maximal dimension of () is dim
= n 1.
Denition 21. The characteristic subset C
x
of T
x
M is dened by
C
x
=
[
C
\
(). (41)
2.2 The Cauchy problem 19
2.2 The Cauchy problem
2.2.1 Cauchy problem for the Einstein equations: local in time, existence and
uniqueness of solutions. In this chapter we shall discuss the work of Y. Choquet-
Bruhat in [8]. This work is based on the reduction of the Einstein equations to wave
equations. To accomplish this reduction one has to introduce harmonic, or wave,
coordinates.
Denition 22. Let (M. g) be a Riemannian manifold. Then a function is called
harmonic if
.
= 0. (42)
where .
= g
(d
).
If the metric g is Lorentzian, then the equation .
= 0 is the wave equation.

Now the problem is the following: Given a coordinate chart (U. .) with . =
(.
0
. .
1
. .
2
. .
3
), nd functions
, j = 0. 1. 2. 3, each of which is a solution of the

wave equation in U,
.
= 0 in U.
and such that, setting
.
(.
0
. .
1
. .
2
. .
3
).
we have a diffeomorphismof the range V in R
4
of the given chart onto another domain
V in R
4
.
U
M
V
V
R
4
.
.
Figure 3
We thus have another chart (U. .) with domain U, i.e. another system of local
coordinates in U. The equation .
= 0 in an arbitrary system of local coordinates

reads
.
= g
d
2
d.
d.
d
d.
= 0. (43)
Suppose now that we use the functions ( .
0
. .
1
. .
2
. .
3
) as the local coordinates in U,
i.e. we express things relative to the new chart (U. .). Setting equal to each one
of the .
for = 0. 1. 2. 3. we have a solution of the above equation. Since

d .
d .
;
=
;
we have
d
2
.
d .
d .
= 0.
So, the equation reads
0 = .
= g
.
Dropping the bars we can say that a system of local coordinates is harmonic if and
only if the connection coefcients in these coordinates satisfy the condition
I
:= g
= 0. (44)
Let us set
I
:= g
.
Then we can write
I
= g
1
2
g
.
Hence the principal part of d
is the following:
P. P. {d
] = g
{d
].
Denoting by 1
the components of the Ricci curvature tensor, let us dene

H
= 1

1
2
(d
). (45)
Then the principal part of H
is
P. P. {H
] =
1
2
g
.
and we have
H
=
1
2
g
T
k2pc
g
k2
d
g
pc
. (46)
where T is a rational function of the metric g of degree 2, the ratio of a homogeneous
polynomial in g of degree 6 to (det g)
2
. Replacing the Einstein equations
1
= 0 (47)
by the reduced (Einstein) equations
H
= 0. (48)
we have a system of non-linear wave equations for the metric components g
.
In the approach of Choquet-Bruhat one studies the Cauchy problem for these
reduced equations. Let us write
1
= H

1
2
S
. (49)
where
S
= d
.
We have
1

1
2
g
1 = H

1
2
g
H
1
2

1
2
g
.
and
V
(S

1
2
g
S

=:

S
J+
) = g
2
V
2

S
(50)
= g
2
(d
2

S
I
k
2
S
k
I
k
2
S
k
). (51)
If we have a solution of the reduced equations, then by virtue of the twice contracted
Bianchi identities
V

1
2
g
= 0. (52)
this solution also satises the equations
V
= 0. (53)
Now, we have S = 2d
with d
= g
2
d
2
. Therefore
= d
d
2
I
2
.
The principal part of V
is
P. P. {V
] = d
(d
) d
(d
2
I
2
)
= g
.
In fact, we have
V
= g
.
where is a linear form in dg with coefcients which are homogeneous rational
functions of g. Therefore the equations
V
= 0
constitute a system of homogeneous linear wave equations for the I
. Consequently
the I
vanish identically provided that the initial conditions vanish, that is

I
0
= 0. (54)
d
0
I
0
= 0. (55)
where
0
is the initial hypersurface .
0
= 0. Given now initial data for the Einstein
equations
1
= 0 or, equivalently,

1
= 0. (56)
where
= 1

1
2
g
1.
we construct initial data
g
0
. (57)
d
0
g
0
(58)
for the reduced equations H
= 0, such that the conditions

I
0
= 0. d
0
I
0
= 0
are satised. Then the solution g
of the Cauchy problem for the reduced equations

shall, according to the above, also satisfy the conditions I
= 0, therefore shall be a
solution of the original Einstein equations.
Initial data for the Einstein equations consist of a pair ( g
i}
. k
i}
) where g
i}
is a
Riemannian metric and k
i}
a 2-covariant symmetric tensor eld on the 3-manifold
M, which is to be identied with the initial hypersurface

0
. Once we have a solution
(M. g) with M = 0. T )
0
and
0
=

M, then g
i}
and k
i}
shall be, respectively,
the 1
st
and 2
nd
fundamental form of
0
= {0]
0
in (M. g). Thus
g
i}
= g
i}
[
0
. i. = 1. 2. 3.
We choose the coordinates to be Gaussian normal along
0
, that is
g
i0
[
0
= 0. (59)
g
00
[
0
= 1. (60)
Then
d
0
g
i}
[
0
= 2k
i}
. (61)
We then choose
d
0
g
0i
[
0
. d
0
g
00
[
0
so as to satisfy the conditions
I
0
= 0.
A short calculation shows that d
0
g
0i
[
0
=

I
i
and d
0
g
00
[
0
= 2 tr k, where

I
i
are the corresponding (3-dimensional) quantities for the induced metric g
i}
. (Re-
call: tr k = g
i}
k
i}
.) This completes the specication of initial data for the reduced
equations. Now consider the following: For a solution of the reduced equations,
1
0i
[
0
=
1
2
S
0i
[
0
=
1
2
{d
0
I
i
d
i
I
0
g
0i
(d
2
I
2
]
[
0
=
1
2
d
0
I
i
[
0
and
1
00
[
0
=
1
2
S
00
[
0
=
1
2
{2d
0
I
0
g
00
(d
2
I
2
]
[
0
=
1
2
d
0
I
0
[
0
.
Therefore, if the initial data ( g
i}
. k
i}
) verify the constraint equations
1
0i
[
0
= 0. (62)
1
00
[
0
= 0. (63)
then the conditions d
0
I
0
= 0 are satised as well.
In the original work of Choquet-Bruhat a local problemwas posed, the initial data
being given on a domain C
0
. As a rst step the initial data for the reduced
equations is extended to the whole of R
3
in such a way that it becomes trivial outside
a larger domain C
t
, where C
t
C, with compact closure in R
3
.
C
C
t
0
Figure 4
The next step in the construction of the solution is based on the domain of depen-
dence theorem, to be formulated below. Let (M. g) be the known spacetime, where
M = 0. T |
0
.
Denition 23. The domain of dependence of Cin the spacetime (M. g) is the subset
of M for which C is a (incomplete) Cauchy hypersurface.
So, the domain of dependence D(C) of C in M is the set of points M such
that each past-directed causal curve in M through intersects C. (It follows that it
cannot intersect more than once.)
0
C
D(C)
Figure 5
In D(C) the solution depends only on the initial data in C. In particular, since the
constraint equations are satised in C, we have that I
and d
0
I
all vanish in C.
So, by the domain of dependence theorem applied to the (linear homogeneous) wave
equations for I
, the I
vanish throughout D(C). Therefore the solution of the

reduced equations is in fact a solution of the Einstein equations
1
= 0 in D(C).
If the 3-manifold

M is compact, one can cover

M with a nite number of coordinate
charts and construct a local time solution by putting together these local solutions.
This works by virtue of the domain of dependence theoremfor the Einstein equations.
For, suppose that C
1
and C
2
are two such coordinate charts with C
1
C
2
= 0.
Since we are given initial data ( g. k) on the whole 3-manifold

M, the representations
of these data, given by the two charts, are related by the diffeomorphismin the overlap.
Thus there exists a diffeomorphism of C
1
C
2
onto itself such that
g
2
=
+
g
1
. k
2
=
+
k
1
.
After making this transformation we may assume that the initial data coincide in
C
1
C
2
. If g
1
and g
2
are the two solutions of the reduced equations, corresponding
to the initial conditions in C
1
and C
2
respectively, the domain of dependence theorem
says that g
1
and g
2
coincide in the domain of dependence of C
1
C
2
relative to
either g
1
or g
2
.
We can therefore extend either solution
(D(C
1
). g
1
). (D(C
2
). g
2
)
to the union, the domain of dependence of C
1
L C
2
.
0
C
1
C
2
Figure 6
Remark. The domain of dependence theorem is a rened uniqueness theorem.
Denition 24. Given initial data (C. g. k) where (C. g) is not required to be com-
plete, we say that a spacetime (U. g) is a development of this data, if C is a Cauchy
hypersurface for U. So Cis the past boundary of Uand g and k are respectively the
rst and second fundamental forms of the hypersurface C in (U. g). Moreover, g
satises the Einstein equations
1
= 0.
An argument analogous to the one just presented shows that if U
1
and U
2
are
developments of the data (C. g. k), then we can dene a development with domain
U
1
L U
2
which extends the corresponding metrics g
1
and g
2
. Therefore, the union
of all developments of given initial data is also a development of the same data, the
maximal development of that data.
Theorem 2 (Y. Choquet-Bruhat, R. Geroch [9]). Any initial data set (
M. g. k)
(completeness of (
M. g) not assumed) satisfying the constraint equations, gives rise

to a unique maximal development.
We shall now give an exposition of the domain of dependence theorem in a gen-
eral Lagrangian setting. Recall the Lagrangian for a mapping u: M N from
Section 2.1.1. Given a background solution u
0
, we dened the quadratic form
h =
J
2
1
J
2
(
0
) with h( . ) = h
ob

o
where h
ob
=
J
2
1

u
J

b
+
and
0
= Ju
0
(.).
Let us denote by {1] the equivalence class of Lagrangians giving rise to the same
EulerLagrange equations. Recall Denition 19 from Section 2.1.1.
Denition 25. We say that {1] is regularly hyperbolic at
0
if the quadratic form h
fullls the following:
1. There is a covector T
+
x
M such that h is negative denite on
1
:= { Q : Q T
q
N] L(T
x
M. T
q
N).
that is, the elements of the form
o
Q
o
.
2. There is a vector X T
x
M with X > 0 such that h is positive denite on
the set
1
A
of rank-1-elements of the subspace
A
= { L(T
x
M. T
q
N) : X = 0].
that is, the elements of the form
o
1
o
where
= 0.
Denition 26. Let h be regularly hyperbolic. Set
1
+
x
= { T
+
x
M : h is negative denite on 1
] (64)
and
J
x
= {X T
x
M : h is positive denite on
1
A
]. (65)
Proposition 2. 1
+
x
and J
x
are open cones each of which has two components
1
+
x
= 1
+
x
L 1
+-
x
.
J
x
= J
x
L J
-
x
.
where 1
+-
x
and J
-
x
are the sets of opposites of elements in 1
+
x
and J
x
, respectively.
Moreover, each component is convex. The boundary d1
+
x
is a component (the inner
component) of C
+
x
, the characteristic in T
+
x
M, and dJ
x
is a component (the inner
component) of C
x
, the characteristic in T
x
M.
Proof. The proof is in the book [12].
Recall the denition of the Noether transform m(. X) of h corresponding to a
pair (. X) T
+
x
M T
x
M with X > 0:
m(. X)(
1
.
2
) = (. X)h(
1
.
2
) h(
1
X.
2
) h(
1
.
2
X).
Proposition 3. Let U
x
T
+
x
M T
x
M be given by
U
x
= {(. X) : X > 0].
Consider the subset of U
x
consisting of those (. X) with m(. X) positive denite
on 1
= { 1 Q : V T
+
x
M. V 1. Q T
q
N]. Then this subset is given
by
(1
+
x
J
x
) L (1
+-
x
J
-
x
).
Moreover, on the boundary of this set m(. X) has nullity.
Let us discuss briey the notion of variation of a mapping u
0
: M N. A
variation of u
0
, namely u, is a section of u
+
0
T N (the pullback by u
0
of T N). In
general, if B is a bundle over N and u
0
: M N, we denote by u
+
0
B the pullback
bundle, namely the following bundle over M:
u
+
0
B =
[
x
{.] B
u
0
(x)
.
where B
q
is the bre of B over q N.
M N
.
u(.) = ,
u
t
(.)
Figure 7
Thus, a variation u maps . M u(.) T
u
0
(x)
N. For a given . M, u(.) is
the tangent vector at u
0
(.) of the curve t u
t
(.) in N, where u
t
is a differentiable
1-parameter family of mappings u
t
: M N, i.e., u(.) =
du
I
(x)
dt
t=0
.
We now explain the meaning of the subset J
x
T
x
M: J
x
is the set of possible
values at . M of a vector eld X on M with the property that the reduced equations
obtained by considering mappings which are invariant under the corresponding 1-
parameter group of diffeomorphisms of M, form a regularly elliptic system.
On the other hand, the subset 1
+
x
T
+
x
M denes the notion of a spacelike
hypersurface.
Denition 27. A hypersurface H in M is called spacelike, if at each . H the
double ray {z : z = 0 R] dened by the hyperplane T
x
H in T
x
M,
T
x
H = {Y T
x
M : Y = 0]. (66)
is contained in 1
+
x
.
Denition 28.

1
x
, the causal subset of T
x
M, is the set of all vectors X T
x
M such
that X = 0 for all 1
+
x
.
1
x
is a closed subset of T
x
M with

1
x
=

1
x
L

1
-
x
, where

1
-
x
is the set of opposites
of elements in

1
x
. One can show that each component is convex. If X d
1
x
then
there exists a covector d1
+
x
such that X = 0. It follows that each component
1
+
x
and 1
+-
x
lies to one side of the plane
A
= { T
+
x
M : X = 0]
and
A
contains a ray of d1
+
x
respectively d1
+-
x
. Thus if d1
+
x
is differentiable at
this double ray, then
A
is the tangent plane to 1
+
x
at this double ray.
Denition 29. A causal curve ; in M is a curve in M whose tangent vector ;(t ) at
each point ;(t ) belongs to

1
;(t)
.
The following statements are valid for the future and past components of

1
x
and J
x
separately. In general, we have

1
x
J
x
. In fact, J
x
is the interior of the inner com-
ponent of C
x
while

1
x
is the convex hull of the outer component of C
x
. (Remember
that C
x
is the characteristic in the tangent space T
x
M.)
1
+
x
T
+
x
M
T
x
M
J
x
1
x
= dual of 1
+
x
Figure 8
We have in general m-sheets in the case dimN = m.
Next, let us give two equivalent denitions of domain of dependence:
Denition 30. Let Rbe a domain in M on which a solution u of the EulerLagrange
equations is dened. Consider a domain D R and a hypersurface in R, which
is spacelike relative to Ju.
1. We say that D is a development of if we can express
D =
[
t0,Tj
t
where {
t
: t 0. T |] is a foliation and where each
t
is a spacelike (relative
to Ju) hypersurface in R homologous to
0
= . (In particular, d
t
= d
for all t 0. T |.)
2. D is a development of if each causal curve in R through any point of D
intersects at a single point.
We sketch the proof of the equivalence of the above two denitions.
Sketch of proof. Suppose that D is a development of according to Denition 30.2.
Then for each point D the causal past J
-
() (we are considering future develop-
ments; is the past boundary of D) is compact. We dene t () to be the volume of
J
-
(), given a volume formc on R. This denes a time function in D\: Jt X > 0
for any vector X

1
x
with . D \ . The level sets of t then dene a foliation
{
t
] as required in Denition 30.1. Conversely, suppose that D is a development of
according to Denition 30.1. Then any causal curve in D can be parametrized by
t in a non-singular manner. It follows that each past-directed causal curve from any
point of D must intersect . One can show that if D
1
and D
2
are developments of
, then D
1
L D
2
is also a development. So given a domain of denition R and a
spacelike hypersurface , we can dene the domain of dependence of in Rrelative
to Ju to be the maximal development.
To state the domain of dependence theorem in a precise and general manner, we
reformulate the general Lagrangian setup from a global perspective. Consider maps
u: M N. The conguration space is C = M N. The velocity space V is a
bundle over C:
V =
[
(x,q)C
L(T
x
M. T
q
N).
We have a projection
: V C.
L(T
x
M. T
q
N) (.. q) C.
where we write
V,
for
1
: V C M and we write
V,N
for
2
: V
C N.
Let us list at this point the relevant notions.
TM. the tangent bundle of M,
i
M. the bundle of (fully antisymmetric) r-forms on M,
S
2
M. the bundle of quadratic (symmetric bilinear) forms on M,
and, with n = dimM,
n
M. the bundle of top-degree-forms on M.
So,
n
M is a bundle over M and
V,
: V M
is the projection dened above. Then the pullback bundle is
+
V,

n
M.
a bundle over V. An element of this is o (
n
M)
x
with . M, and this o is
attached to an element L(T
x
M. T
q
N). The Lagrangian 1 is actually a C
o
-
section of
+
V,

n
M over V. So,
1() (
n
M)
x
: L(T
x
M. T
q
N).
Thus
1()(Y
1
. . . . . Y
n
).
with Y
1
. . . . . Y
n
T
x
M, is an n-linear fully antisymmetric form on T
x
M.
Denition 31. Given a Lagrangian 1, the action of a map u dened in a domain R
in M, corresponding to a subdomain D R is
Au: D| =
Z
D
1 Ju. (67)
Note that 1 Ju is a section of
n
M over R ((1 Ju)(.) = 1(Ju(.))
(
n
M)
x
). Also, note that with dimM = n, dimN = m, we have dimC = n m,
dimV = (n m) nm.
Suppose that M is oriented and c is a C
o
volume form on M. That is, c is a
C
o
-section of
n
M such that if (1
1
. . . . . 1
n
) is a positive basis for T
x
M (a positive
frame at .), then c(1
1
. . . . . 1
n
) > 0. Given then a C
o
function 1
+
on V, we dene
the corresponding Lagrangian 1 by
1()(Y
1
. . . . . Y
n
) = 1
+
() c(Y
1
. . . . . Y
n
)
with L(T
x
M. T
q
N) and Y
1
. . . . . Y
n
in T
x
M. The function 1
+
was called
Lagrangian in the previous.
Finally, we state the domain of dependence theorem.
Theorem 3 (Domain of dependence theorem). Let u
0
be a C
2
solution of the Euler
Lagrange equations corresponding to a C
o
Lagrangian 1; and let u
0
be dened in
a domain Rin M. Let be a hypersurface in R, which is spacelike relative to Ju
0
.
Let also u
1
be another solution of the EulerLagrange equations dened and C
1
on R. Suppose that
Ju
0
[
= Ju
1
[
.
Then u
1
coincides with u
0
in the domain of dependence of in R relative to Ju
0
.
2.3 Decomposition of the Einstein equations 31
2.3 Decomposition of the Einstein equations with respect to the
foliation by the level sets H
t
of a time function t
We consider rst Lorenzian geometry without the Einstein equations, and afterwards
we impose the Einstein equations.
H
0
H
t
Figure 9
Let (1
1
, 1
2
, 1
3
) be a (local) frame eld for H
0
. We extend the 1
i
, i = 1. 2. 3,
to the spacetime by the condition
T. 1
i
| = 0. (68)
Then the (1
1
. 1
2
. 1
3
) dene a frame eld for each H
t
. The spacetime manifold M
is represented as the product 0. T | H
0
. In this representation T =
J
Jt
and
g =
2
Jt
2
g.
where g(t ) is the induced metric on H
t
and is the lapse function. Recall the rst
variation equations
d g
i}
dt
= 2k
i}
. k
i}
= k(1
i
. 1
}
). (69)
Here k is the second fundamental form of H
t
and the indices i , refer to the frame
(1
i
: i = 1. 2. 3). The second variation equations are
dk
i}
dt
=

V
i
V
}
(1
i0}0
k
in
k
n
}
). (70)
Here

V is the covariant derivative operator intrinsic to H
t
, that is, relative to the
Riemannian connection of (H
t
. g(t )). Also, 1
i0}0
= 1(1
i
. 1
0
. 1
}
. 1
0
), the frame
eld (1
1
. 1
2
. 1
3
) being completed by 1
0
=
1
J
Jt
, the future-directed unit normal to
the H
t
, to a frame eld for M. Remark that g
00
= 1, g
0i
= 0 and g
i}
= g
i}
=
g(1
i
. 1
}
).
In addition to the 1
st
and 2
nd
variation equations we have the Codazzi and Gauss
equations of the embedding of H
t
in M. The Codazzi equations are
V
i
k
}n

V
}
k
in
= 1
n0i}
. (71)
and the Gauss equations are
1
in}n
k
i}
k
nn
k
i n
k
n}
= 1
in}n
. (72)
Here

1
in}n
are the components of the curvature tensor of (H
t
. g(t )).
We proceed to impose the Einstein equations. Taking the trace of the Gauss
equations we obtain
1
i}
tr k k
i}
k
in
k
n
}
= 1
i}
1
i0}0
. (73)
Next we substitute for 1
i0}0
from these equations into the 2
nd
variation equations to
conclude that the part 1
i}
= 0 of the vacuum Einstein equations is equivalent to
dk
i}
dt
=

V
i
V
}
(
1
i}
k
i}
tr k 2k
in
k
n
}
). (74)
The trace of the Codazzi equations is
V
i
k
i}
d
}
tr k = 1
0}
. (75)
The part 1
0}
= 0 of the vacuumEinstein equations is thus equivalent to the constraint
equation
V
i
k
i}
d
}
tr k = 0. (76)
The double trace of the Gauss equations is
1 (tr k)
2
[k[
2
= 1 21
00
= 2

1
00
. (77)
where [k[
2
= k
i
n
k
n
i
and

1
= 1
1
2
g
1. Thus the vacuum Einstein equation
1
00
= 0 is equivalent to the constraint equation
1 (tr k)
2
[k[
2
= 0. (78)
The constraint equations constrain the initial data
( g. k) on H
0
.
To derive the 2
nd
variation equations we must obtain an expression for the accel-
eration of the orthogonal family of curves.
2.3 Decomposition of the Einstein equations 33
Acceleration of the integral curves of T . Let us denote the unit normal 1
0
= N.
Then the geodesic curvature of the orthogonal family of curves is given by
V
1
N =
-1
V. (79)
and

V is the gradient of intrinsic to H
t
, a vector eld tangent to H
t
with com-
ponents
V
i
= g
i}
V
}
.
where

V
}
= 1
}
. Recall that T = N (the integral curves of T and N are
the same, but differ only by parametrization; T is parametrized by t while N is
parametrized by s, namely arc length.) The formula (79) is derived as follows. We
have (in arbitrary local coordinates)
N
= g
= d
t. (80)
Hence, we can write
N
= N
(d
t )
= N
t N
(d
t )
=
-1
N
(
-1
N
).
The last term is
-1
N
=-1
d

=
1
2
J
J
(N
)

D1
=0
=
-1
d
.
Thus,
N
=
-1
.
where
= g
= g
2
2
and
2
denes the orthogonal projection
to the H
t
:
X = X g(N. X)N
on any vector X TM. We have thus obtained the formula (79). Since T = N,
an expression for the acceleration V
T
T of the integral curves of T readily follows.
Now let X, Y be vector elds tangential to the H
t
and satisfying
T. X| = T. Y | = 0.
At the end we shall set X = 1
i
and Y = 1
}
to obtain T k(1
i
. 1
}
) =
Jk
i
Jt
. We have
k(X. Y ) = g(V
A
N. Y )
=
-1
g(V
A
T. Y ).
Therefore
T(k(X. Y )) = T(
-1
g(V
A
T. Y )).
Now,
T(g(V
A
T. Y )) = g(V
T
V
A
T. Y ) g(V
A
T. V
T
Y ).
For the rst term we apply the denition of the curvature:
V
T
V
A
T = V
A
V
T
T V
T,Aj
T 1(T. X)T
and
g(Y. 1(T. X)T ) = 1(Y. T. T. X) = 1(X. T. Y. T ).
For the second term we use
V
T
Y V
Y
T = T. Y | = 0.
to write
g(V
A
T. V
T
Y ) = g(V
A
T. V
Y
T ).
Substituting T = N, we can express the vector eld 7 := V
A
N in terms of k. In
fact 7 is tangential to the H
t
, hence 7 = 7
i
1
i
and 7
i
g
i}
= g(7. 1
}
) = k(X. 1
}
).
The 2
nd
variation equations (70) followin this manner after substituting for V
T
T from
the formula (79) for V
1
N.
To derive the Gauss equations, we recall from Riemannian geometry that the
covariant derivative

V intrinsic to (H
t
. g(t )) is characterized by the property
V
A
Y = V
A
Y.
where X, Y are any vector elds tangential to the H
t
. We apply this to the denition
of the curvature of (H
t
. g(t )). Let X. Y. 7 be arbitrary vector elds tangential to H
t
.
Then
1(X. Y )7 =

V
A
V
Y
7

V
Y

V
A
7

V
A,Y j
7. (81)
Now let W be another arbitrary vector eld tangential to H
t
. Then we can write
1(W. 7. X. Y ) = g(W.

1(X. Y )7) (82)
= g(W. V
A
V
Y
7 V
Y

V
A
7 V
A,Y j
7). (83)
since g(W. U) = g(W. U) for any vector eld U. At this point we substitute
V
A
7 = V
A
7 = V
A
7 g(N. V
A
7)N.
Thus in V
Y
(
V
A
7) there are terms, in addition to V
Y
V
A
7, which involve V
Y
N, thus
the second fundamental form k. The Gauss equations (72) follow in this manner.
The Codazzi equations (71) are straightforward.
3 Asymptotic atness at spacelike innity and
conserved quantities in General Relativity
3.1 Conserved quantities
In this chapter we discuss the denitions of total energy, linear momentumand angular
momentum in General Relativity. We then give an overview followed by a rigorous
discussion of the associated conservation laws.
We begin with the denitions of a manifold which is Euclidean at innity, of an
asymptotically Euclidean Riemannian manifold and of an asymptotically at initial
data set.
Denition 32. A 3-manifold H is said to be Euclidean at innity if there exists a
compact set K H such that H \ K is diffeomorphic to R
3
\ B, where B is a ball
in R
3
. Thus H \ K is contained in the domain of a chart.
Denition 33. An asymptotically Euclidean Riemannian manifold (H. g) is a com-
plete Riemannian manifold which is Euclidean at innity and there exists a coordinate
system(.
1
. .
2
. .
3
) in the complement of K above relative to which the metric com-
ponents g
i}

i}
as r :=
q
P
3
i=1
(.
i
)
2
o.
Denition 34. An asymptotically at initial data set (H. g. k) is an initial data set
where (H. g) is an asymptotically Euclidean Riemannian manifold and the compo-
nents of k approach 0 relative to the coordinate system above as r o.
The fall-off of g
i}

i}
and k
i}
with r should be sufciently rapid for the notions
of total energy, linear momentum and angular momentum below to be well dened
and nite.
The standard denition of these notions is the following:
Denition 35 (Arnowitt, Deser, Misner (ADM) [1]). Let S
i
= {[.[ = r] be the
coordinate sphere of radius r and JS
}
the Euclidean oriented area element of S
i
. We
then dene
the total energy
1 =
1
4
lim
i-o
Z
S
X
i,}
(d
i
g
i}
d
}
g
i i
) JS
}
. (84)
36 3 Asymptotic atness at spacelike innity and conserved quantities
the linear momentum
1
i
=
1
2
lim
i-o
Z
S
(k
i}
g
i}
tr k) JS
}
. (85)
the angular momentum
J
i
=
1
2
lim
i-o
Z
S
c
i}n
.
}
(k
nn
g
nn
tr k) JS
n
. (86)
The total energy, linear momentum and angular momentum are conserved quan-
tities at spacelike innity, as shall be shown in the sequel.
We now begin the discussion of where these quantities come from. Let us recall
the fundamental theorem of Noether.
Theorem 4 (Noethers theorem [22]). In the framework of a Lagrangian theory, to
each continuous group of transformations leaving the Lagrangian invariant there
corresponds a quantity which is conserved.
In particular:
energy corresponds to time translations,
linear momentum corresponds to space translations,
angular momentum corresponds to space rotations.
As discussed in the previous chapter, a Lagrangian corresponds in a local coordi-
nate description to a function
1
+
= 1
+
(.. q. ).
We denote by .
(with j = 1. . . . . n) the independent variables, q

o
= u
o
(.) (with
a = 1. . . . . m) the dependent variables, and
o
=
Ju
u
Jx
J
(.) the rst derivatives of the
dependent variables. The canonical momentum is given by
+
o
=
d1
+
d
o
. (87)
We dene the canonical stress by
T
+
=
+
o

o
1
+
. (88)
In the following we restrict ourselves to transformations acting only on the domain
of the independent variables. Let X
be a vector eld generating a 1-parameter group

of transformations of this domain leaving invariant the Lagrangian form
1
+
J
n
.. J
n
. = J.
1
. . J.
n
. (89)
3.1 Conserved quantities 37
Then the Noether current
J
+
= T
+
(90)
is divergence-free, that is
d
J
+
= 0. (91)
By the divergence theorem we conclude that the following conservation law holds:
If
1
and
2
are homologous hypersurfaces (in particular d
1
= d
2
), then
Z
2
J
+
J
=
Z
1
J
+
J
. (92)
We now turn to General Relativity. The Einstein equations are derived from
Hilberts actionprinciple. This corresponds to the following Lagrangian for gravity:
1 =
1
4
1 Jj
.
where 1 is the scalar curvature of the metric g and Jj
the volume form of g. This

Lagrangian is the only one (up to an additive constant multiple of the volume form)
which gives rise to second order (EulerLagrange) equations. Here the metric com-
ponents g
are the unknown functions and 1 depends on their second derivatives.

Moreover, it is the only geometric invariant (up to an additive constant) which con-
tains the second derivatives only linearly. (All other invariants give rise to 4
th
order
equations.) But Noethers theorem depends on having a Lagrangian containing only
the rst derivatives of the unknown functions. Thus it cannot be applied directly to
Hilberts principle.
To deal with this difculty, Einstein and Weyl introduced the Lagrangian
1
4
1 d
= 1
+
.
where 1
=
1
4
_
g (g
). Then
1
+
=
1
4
_
g g
(I
)
differs from Hilberts invariant Lagrangian
1
4
1
_
g by a divergence, therefore
gives rise to the same eld equations. So, one can dene
+
=
d1
+
d
to be the canonical momentum, where
= (d
)(.) (q
o
corresponding to
g
(.)). The canonical stress is then

T
+
=
+
1
+
.
This is called the Einstein pseudo-tensor. The Lagrangian 1
+
is invariant under
translations: .
, where the c
are constants. Let

J
+
= T
+
. where X
= c
(X = c
d
d.
).
so X is a vector eld generating a 1-parameter group of translations. By Noethers
theorem the current J
+
is divergence free, that is
d
J
+
= 0.
Since the c
are arbitrary constants, the following differential conservation laws hold:

d
T
+
= 0.
Hermann Weyl wrote the following comments about these conservation laws. We
quote from the book [29], p. 273:
In the original German:
Dennochscheint es physikalischsinnlos zusein, die T
+
als Energiekom-
ponenten des Gravitationsfeldes einzufhren; denn diese Grssen bilden
weder einen Tensor noch sind sie symmetrisch. In der Tat knnen durch
geeignete Wahl eines Koordinatensystems alle T
+
an einer Stelle stets

zum Verschwinden gebracht werden; man braucht dazu das Koordi-
natensystem nur als ein geodtisches zu whlen. Und auf der andern
Seite bekommt man in einer Euklidischen, vllig gravitationslosen
Welt bei Benutzung eines krummlinigen Koordinatensystems T
+
, die
verschieden von 0 sind, wo doch von der Existenz einer Gravitations-
energie nicht wohl die Rede sein kann. Sind daher auch die Differential-
relationen (oben) ohne wirkliche physikalische Bedeutung, so entsteht
doch aus ihnen durch Integration ber ein isoliertes System ein invari-
anter Erhaltungssatz.
In English translation:
Nevertheless it seems to be physically meaningless to introduce the T
+
as energy components of the gravitational eld; for, these quantities

are neither a tensor nor are they symmetric. In fact by choosing an
appropriate coordinate system all the T
+
can be made to vanish at

any given point; for this purpose one only needs to choose a geodesic
(normal) coordinate system. And on the other hand one gets T
+
= 0
in a Euclidean completely gravitationless world when using a curved
coordinate system, but where no gravitational energy exists. Although
the differential relations (d
T
+
= 0, above) are without a physical

meaning, nevertheless by integrating them over an isolated system one
gets invariant conserved quantities.
Consider the transformation
. . = (.).
g
+
g.
In components,
.
=

(.)
and
g
. g
(.) =
d

d.
d

d.
( .)
(. =
-1
( .)). As an example we have, in particular, linear transformations
.
= a
.
where a and b are constants. Then we have
g
(.) = a
( .).
If {
t
] is a 1-parameter group of transformations generated by a vector eld X, then
J
Jt
+
t
g
t=0
= L
A
g (93)
is the Lie derivative of g with respect to X. We have
(L
A
g)
= V
.
where X
= g
. We say that an integrated quantity Q is gauge invariant if for

every such 1-parameter group {
t
],
Q
+
t
g| = Qg|. (94)
This requirement implies
Q =
J
Jt
Q
+
t
g|
t=0
= 0 (95)
and conversely. In view of (93), the last reads
Q := D
Q L
A
g = 0. (96)
Here, we think of Q as a differentiable function of g.
Now, the metric g is not a mapping of the (spacetime) manifold M into another
manifold N, but rather a section of a tensor bundle over M. Thus, Noethers theorem
must be extended to sections of tensor bundles. In any case, we have derived, by
considering translations,
d
T
+
= 0. (97)
Let us now consider rotations. The vector elds
X
(i})
= .
i
d
d.
}
.
}
d
d.
i
. i. = 1. 2. 3.
are the generators of rotations in the (i. )-plane. In 3-space-dimensions the vector
elds
X
(i)
= c
i}k
.
}
d
d.
k
generate rotations about the i
th
coordinate axis. In Minkowski spacetime the vector
eld
X
()
= .
d
d.
d
d.
. .
= j
.
generates spacetime-rotations in the (.
. .
)-coordinate plane. Here j =

diag(1. 1. 1. 1) is the Minkowski metric in rectangular coordinates. We denote
the vector eld generating translations along the -coordinate axis by
X
()
=
d
d.
.
Now, Noethers theorem in the case of translations applies as it stands if we take the
naive point of view of considering g as a matrix-valued function on M = R
4
. This
yields
d
J
+
()
= 0. where J
+
()
= T
+
()
= T
+
.
On the other hand, for spacetime rotations in the (. )-plane we have
J
+
()
= T
+
()
= .
T
+
T
+
and
d
J
+
()
= T
+
T
+
.
noting that d
= j
= j
. The last vanishes if and only if the matrix with

entries
S
= j
T
+
is symmetric, that is S
= S
. But this is false. Therefore the above argument

does not yield a conservation law corresponding to spacetime rotations. This is not
surprising: The naive point of view fails for rotations, because rotations, in contrast
to translations, bring the tensor character of g into play.
We shall presently discuss Noethers theorem from a global geometric point of
view. Consider maps
u: M N
with dimM = n and dimN = m. As in Section 2.2, we denote by V the velocity
bundle
V =
[
(x,q)
L(T
x
M. T
q
N).
where u(.) = q N and Ju(.) = L(T
x
M. T
q
N). A diffeomorphism of
M onto itself induces a diffeomorphism
+
of V onto itself by
L(T
x
M. T
q
N)
+
() L(T
((x)
M. T
q
N). (98)
where
+
() Y = (J
-1
Y ) V Y T
((x)
M. (99)
If X is a vector eld on M generating the 1-parameter group {
t
], we call the induced
1-parameter group {
t+
] of diffeomorphisms of V the Lie ow generated by X on V.
Recall that a Lagrangian 1is a section of the bundle
+
V,
n
M over V, the pullback
by the projection
V,
: V M of the bundle
n
M of top-degree-forms on M.
The pullback
+
by of 1 is dened by
(
+
1)() (Y
1
. . . . . Y
n
) = 1(
+
()) (J Y
1
. . . . . J Y
n
) (100)
for all L(T
x
M. T
q
N) and all Y
1
. . . . . Y
n
T
x
M. (Recall that
+
()
L(T
((x)
M. T
q
N).) The Lie derivative L
A
1 of 1 with respect to a vector eld X on
M is the derivative of 1 with respect to the Lie ow generated by X on V:
L
A
1 =
J
Jt
+
t
1
t=0
. (101)
where {
t
] is the 1-parameter group generated by X. A current J is a section of
+
V,
n-1
M. Given a volume form c on M we consider J as equivalent to J
+
, a
section of
+
V,
TM, as follows:
J() (Y
1
. . . . . Y
n-1
) = c(J
+
(). Y
1
. . . . . Y
n-1
). (102)
for all L(T
x
M. T
q
N) and all Y
1
. . . . . Y
n-1
T
x
M. We now refer to the
following notion from [12].
Denition 36. Given a Lagrangian 1, we say that a current J is compatible with
1, if there exists a section 1 of
+
V,
n
M such that for every solution u of the
EulerLagrange equations corresponding to 1 we have
J (J Ju) = 1 Ju. (103)
In the above denition J Ju is an (n1)-form on M while 1Ju is an n form
on M.
An interesting question is the following: What are all the possible currents com-
patible with a given Lagrangian? In n = 1 dimension all currents are compatible.
However, the question becomes non-trivial in n > 1 dimensions. In fact, the Noether
theorem provides a class of compatible currents in any dimension n of the domain
manifold M.
Denition 37. The domain Noether current corresponding to a Lagrangian 1 and a
vector eld X on M is the current J, given by
J
+
= T
+
. (104)
where T
+
are the components of the canonical stress. Recall that T

+
=
+
o

o

1
+
and 1() = 1
+
()c(.) while
+
o
=
J1
J
u
J
are the components of the canonical
momentum.
We can now state Noethers theorem in the domain case.
Theorem 5 (Noethers theorem in the domain case). The Noether current is a com-
patible current and the corresponding section 1 of
+
V,
n
M is given by
1 = L
A
1. (105)
In particular, if X generates a Lie ow leaving the Lagrangian invariant, then J is
conserved, that is for every solution u of the EulerLagrange equations we have
J(J Ju) = 0. (106)
Hence, for two homologous hypersurfaces
1
and
2
it holds that
Z
2
J Ju =
Z
1
J Ju. (107)
Sections of Tensor Bundles over M. Having in mind the application to General
Relativity we consider the case of the bundle S
2
M of 2-covariant symmetric tensors
on M. In fact, the conguration space C in General Relativity is the open subbundle
1M of Lorentzian metrics on M:
C = 1M =
[
x
1
x
M. (108)
where 1
x
M is the set of quadratic forms on T
x
M of index 1, an open subset of the
space S
2x
M of quadratic forms on T
x
M. The velocity bundle V is
V =
[
x
L(T
x
M. S
2x
M).
The Lagrangian is dened on the bundle product
C
V =
[
x
1
x
M L(T
x
M. S
2x
M).
(Note that dim(C
V) = dimC dimV dimM = n

n(n1)
2

n
2
(n1)
2
.)
Given a diffeomorphism of M onto itself we shall dene the action of on C,
V and on C
V. Consider a section s of C (over M). Thus, s is a Lorentzian

metric on M. We must associate to s a derived section which is a section of V (over
M). Here, a connection I (on TM) is needed. We assume that I is symmetric.
The derived section is then Vs, the covariant derivative of s with respect to I. To
represent things in terms of duals, we must also choose a volume form c on M. This
must be compatible with I, that is we must have
Vc = 0.
Such a choice is possible if and only if
tr 1(X. Y ) = 0 V X. Y T
x
M. V. M.
where 1(X. Y ) is the curvature transformation associated to I. (In an arbitrary local
frame this condition reads 1
= 0.) We now give the following denitions:

Denition 38. The action of on C is dened by
q C
x

+
(q) C
((x)
. . M. (109)
where
+
(q) (Y
1
. Y
2
) = q (J
-1
Y
1
. J
-1
Y
2
) for all Y
1
. Y
2
in T
((x)
M.
Denition 39. The action of on V is dened by
V
x
= L(T
x
M. S
2x
M)
+
() V
((x)
= L(T
((x)
M. S
2((x)
M). . M.
(110)
where
+
() Y =
+
( (J
-1
Y )) for all Y in T
((x)
M.
The Lagrangian 1 is a section of
+
C
^
V,
n
M.
Thus, if (q. ) 1
x
M L(T
x
M. S
2x
M), then
1(q. )
nx
M.
where
nx
M is the space of totally antisymmetric n-linear forms in T
x
M. The
pullback by of 1, namely
+
1, is given by
(
+
1)(q. ) (Y
1
. . . . . Y
n
) = 1(
+
(q).
+
()) (J Y
1
. . . . . J Y
n
) (111)
for all Y
1
. . . . . Y
n
in T
x
M. If X is a vector eld on M generating the 1-parameter
group {
t
] of diffeomorphisms of M onto itself, the induced 1-parameter group {
t+
]
of diffeomorphisms of C
V onto itself , is the Lie owgenerated by X on C
V.
It is actually generated by a vector eld X
+
on C
V which is expressed as the

sum of its horizontal and vertical parts:
X
+
= X
1
+
X
V
+
. (112)
We have
J
C
^
V,
X
1
+
= X (113)
and
J
C
^
V,
X
V
+
= 0. (114)
The horizontal part X
1
+
= X
[
is the horizontal lift of X to C
V dened by the
connection I. The vertical part is given by
X
V
(q. ) =
V
t+
(q)
Vt
t=0
.
V
t+
()
Vt
t=0
.
(an element of S
2x
M L(T
x
M. S
2x
M)) which is a vector tangent to the bre
C
x
L(T
x
M. S
2x
M), an open set in the linear space S
2x
M L(T
x
M. S
2x
M).
1
Here,
t+
(q) is a eld of Lorentzian tensors along the curve
t
(.), namely the inte-
gral curve of X through ..
.
q
(q)
M
C
t
(.)
Figure 10
1
Recall that a tangent vector at a point in a linear space can be thought of as an element of the linear
space (a tangent vector at the origin).
Consider an arbitrary system of local coordinates on M. We can expand q C
x
and L(T
x
M. S
2x
M) in this system as
q = q
J.
(.) J.
(.).
where q
= q
, so q is symmetric, non-degenerate and of index 1. Moreover we

can expand
=
J.
(.) J.
(.) J.
(.).
where
. Given X T
x
M, we have X = X
J
Jx
x
and
X = ( X)
J.
(.) J.
(.).
( X)
.
The (q
) constitute a system of linear coordinates for C

x
and the (
) constitute
a system of linear coordinates for V
x
. We can then express
X
V
+
(q. ) =
V
t+
(q)
Vt
t=0
d
dq
V
t+
()
Vt
t=0
d
d
.
where
V
t+
(q)
Vt
t=0
= q
;
V
X
;
(.) q
;
V
X
;
(.).
V
t+
()
Vt
t=0
(.)
(.)
(.).
In conclusion, for any differentiable function J on C
V we have
X
+
J = X
[
J (q
;
V
X
;
q
;
V
X
;
)
dJ
dq
)
dJ
d
.
The derivative of 1 with respect to the Lie ow generated by X is
L
A
1 =
J
+
t
1
Jt
t=0
. (115)
where 1 =
1
n
P
1
2
1
...2
n
J.
2
1
. . J.
2
n
. We have
(L
A
1)
2
1
...2
n
= X
+
(1
2
1
...2
n
)
n
X
i=1
dX
k
d.
2
i
1
2
1
~2
i
>k...2
n
.
Here, by < z
i
> k we mean that in the i
th
place the sufx z
i
is missing and in its
place we have the sufx k. Writing 1 = 1
+
c, where c is a differentiable volume
form on M (Vc = 0), we have
(L
A
1)
+
= X
+
(1
+
) (V
)1
+
. (116)
We now give the following denition.
Denition 40. In a theory of Lorentzian metrics, the Noether current corresponding
to a vector eld X is
J
+
(q. ) =
+
(X
) 1
+
X
. (117)
The EulerLagrange equations are, in terms of the canonical momentum
+
=
J1
J
J
and the canonical force
+
=
J1
Jq
,
V
(
+
(s. Vs)) =
+
(s. Vs). (118)
We then compute
V
(J
+
(s. Vs))
= V
(
+
(s. Vs)) (X
+
(s. Vs) {X
(V
)(V
) (V
)(V
) (V
)(V
)]
X
[
(1
+
)
d1
+
dq
(s. Vs)V
A
s
d1
+
d
(s. Vs)V
A
(V
) (V
)1
+
.
and we have
(L
A
1)
+
= X
[
(1
+
)
d1
+
dq
(q
d1
+
d
) (V
)1
+
.
Hence, by the denitions of canonical momentum and force as well as the Euler
Lagrange equations,
V
(J
+
(s. Vs)) (L
A
1)
+
(s. Vs)
=
+
(s. Vs) {X
V
A
V
].
(119)
We have
X
(V
) = X
(1
k
s
k
1
k
s
k
).
Let us recall that
V
X
2
1
2
k
X
k
= (L
A
I)
2
.
Consequently, the right-hand side of (119) is
+
(s. Vs) {s
(L
A
I)
(L
A
I)
]. (120)
In fact, the Lie derivative of a connection I is dened by:
(L
A
I) (Y. 7) = X. V
Y
7| V
A,Y j
7 V
Y
X. 7| (121)
for any three vector elds X, Y , 7. One can readily check that (L
A
I) (Y. 7) is
bilinear in Y and 7 with respect to multiplication by the ring of differentiable func-
tions. (That is, we have (L
A
I)( Y. 7) = (L
A
I) (Y. 7) and (L
A
I)(Y. 7) =
(L
A
I) (Y. 7), for any differentiable function .) Moreover, the fact that I is a
symmetric connection implies that (L
A
)I (Y. 7) is symmetric in Y. 7. It follows
that L
A
I is a T
1
2
tensor eld which is symmetric in the lower indices. Since we have
U. V | = V
U
V V
V
U for any pair of vector elds U. V , (121) is equivalent to
(L
A
I) (Y. 7) = V
A
V
Y
7 V
Y
7
X V
A,Y j
7 V
Y
V
A
7 V
Y
V
7
X
= 1(X. Y )7 V
2
X (Y. 7).
(122)
Substituting a frame eld, we obtain
(L
A
I)
2
= 1
2k
X
k
V
V
2
X
. (123)
in agreement with the formula above.
We conclude from the above that the following form of Noethers theorem holds
for sections of C:
Theorem 6 (Noethers theorem for sections of C). The Noether current (from De-
nition 40) is a compatible current, that is
J (J (s. Vs)) = 1 (s. Vs). (124)
and we have
1 = L
A
1 T. (125)
where
T
+
=
{q
(L
A
I)
(L
A
I)
]. (126)
In particular, if X generates a ow on M leaving I invariant as well as a Lie ow
on C
V leaving 1 invariant, then for every solution s of the EulerLagrange

equations we have
J (J (s. Vs)) = 0. (127)
We are now going to apply the above to General Relativity.
We introduce a background metric

g
(in addition to the actual metric g). The
background metric is not to be varied in the variational principle. To

g
is associated
its metric connection

I
and volume formc = Jj
(which is of course

I
-compatible).
We write
1
4
(1 g

1
) Jj
= (1
+

V
1
) Jj
. (128)
Dening the positive function o by
Jj
= o Jj
. (129)
1
+
is given by
1
+
=
1
4
o g
(.
). (130)
where .is the difference of the two connections:
.
= I
=
1
2
g
V
g

V
g
).
(131)
Now 1
+
is a true Lagrangian, as it depends only on g and

V
g. Moreover,
1
=
1
4
o (g
). (132)
The EulerLagrange equations corresponding to 1
+
coincide with the Einstein equa-
tions, provided that

g
is itself a solution of the Einstein equations, that is
1
= 0. (133)
We take

g
to be at from this point on. The background spacetime is then the
Minkowski spacetime. Let the vector eld X generate a 1-parameter group of isome-
tries of

g
. Then X leaves

I
, as well as Jj
= c, invariant. Moreover, the Lie ow

generated by X leaves 1
+
invariant,
X
+
(1
+
) = 0. (134)
We can thus apply Noethers theorem to conclude that we have a conserved current
J
A
associated to X.
Now, the background Minkowski spacetime has a 10-parameter isometry group,
the Poincar group. In Minkowski spacetime there is a special preferred class of
coordinates, the rectangular coordinates, in which the metric components are
= j
= diag(1. 1. 1. 1). (135)

and
Jj
= J.
0
. J.
1
. J.
2
. J.
3
.
Moreover, it holds (as in all linear coordinates) that
= 0. (136)
The generators of the Poincar group are expressed in rectangular coordinates by
X
()
=
d
d.
generators of spacetime translations

along the .
-coordinate axis,
X
()
= .
d
d.
d
d.
generators of spacetime rotations in the

(.
. .
)-coordinate plane, where .
= j
.
In terms of components in rectangular coordinates,
X
()
=
.
X
()
= .
.
We therefore have the conserved Noether currents:
J
()
associated to the translations X
()
,
J
()
associated to the rotations X
()
.
Substituting in Denition 40 the generators of translations along the .
-coordinate
axis we obtain
J
+
()
= T
+
()
. (137)
The associatedconservationlaws coincide withthe EinsteinWeyl energy-momentum
conservation laws, discussed in the preceding. But whereas the EinsteinWeyl ap-
proach provides no conservation lawcorresponding to spacetime rotations, the present
approach does.
Consider the Noether current J
A
associated to a vector eld X generating a 1-
parameter group of isometries of the background Minkowski metric

g
. We have
shown above that J
A
is conserved. This means that for every solution g of the
EulerLagrange equations (the Einstein equations) we have
J (J
A
(g.
V
g)) = 0. (138)
which implies that if
1
and
2
are homologous hypersurfaces (in particular d
2
=
d
1
), then the following holds:
Z
2
J
A
(g.
V
g) =
Z
1
J
A
(g.
V
g). (139)
Next we are going to showthat in fact J
A
is a boundary current. In the general setting
of a Lagrangian theory of maps u of the domain manifold M into another manifold
N, we have the following denition.
Denition 41. A current J is called a boundary current if there exists a section G
of
+
V,
n-2
M (with dimM = n), such that for every solution u of the Euler
Lagrange equations we have
J Ju = J (G Ju). (140)
A boundary current is a conserved current; in fact
Z
J Ju =
Z
J
G Ju. (141)
The integral conservation law
Z
2
J Ju =
Z
1
J Ju
trivially follows from the fact that d
2
= d
1
whenever
2
is homologous to
1
.
Let us go back to the denition of the Noether current corresponding to a vector
eld X in a Lagrangian theory of Lorentzian metrics on M. In rectangular coordinates
of the background Minkowski metric

g
,
= j
V
= d
. det

g
= 1.
and we have
J
+
A
= T
+
+
(g
). (142)
Taking X to be a Killing eld of

g
, we have
X
=
8
<
:
X
()
=
for the generators of spacetime

translations,
X
()
= .
. .
= j
for the generators of spacetime

rotations.
It holds then that
d
= 0.
which is the equation
L
A
I
= 0.
Now, in General Relativity it turns out that we have, identically,
T
+
= d
2
J
+2

1
2
p
det g

1
. (143)
where J
+2
is an expression which is antisymmetric in j and z, to be given below.

(Recall that

1
= 1

1
2
g
1 and

1
= g
2

1
2
.) Fixing the index v let us
write
T
();
= T
+
c
;
(144)
and
J
()
=
1
2
J
+2
c
2
. (145)
Then by (143) for every solution g of the Einstein equations we have
T
()
= J J
()
. (146)
when composed with (g. dg). The J
+2
are given by
J
+2
=
2k
d
k
g
. (147)
where
2k
=
_
det g
8
{
(g
k
g
2
g
2k
g
(g
k
g
2
g
2k
g
)
2g
g
2k

2
g
k
) g
k
(g
g
2
) (148)
g
k
(g
g
2
)].
Using (143) we deduce that for any Killing eld X of

g
and every solution of the
Einstein equations we have
J
+
A
= d
2
G
+2
A
. (149)
where
G
+2
A
= J
+2
1
+2
A
.
and 1
+2
A
is an expression which is antisymmetric in j and z:
1
+2
A
=
1
4
{ (
_
g g
k
j
k
) d
k
X
2
(
_
g g
2k
j
2k
) d
k
X
].
In fact we have the identity
J
+
A
= d
2
G
+2
A

1
2
p
det g

1
. (150)
Thus, for any Killing eld X of

g
and every solution of the Einstein equations we
have
J
A
= JG
A
.
where
(G
A
)
=
1
2
G
+2

c
2
.
We shall now investigate the invariance properties of integrated quantities under
gauge transformations, to be explained presently. Remark that we are free to pull-
back the metric g by an arbitrary (orientation preserving) diffeomorphism of M
while keeping the background Minkowski metric

g
xed:
g
+
g.
where
(
+
g)
(.) =
d

d.
(.)
d

d.
(.)g
((.)).
We consider the quantity Q associated to X, a given Killing eld of

g
, and to S, a
2-surface which is homologous to 0 (S = d for some bounded hypersurface ):
Q
A
(S) =
Z
S
G
A
. (151)
with G
A
= G
A
(g. dg) dened above. Here, S and X are part of the background
Minkowski structure. They are not affected by . So, acts like a gauge transfor-
mation. Consider then a 1-parameter group {
t
] generated by a vector eld Y .
g =
J
Jt
+
t
g
t=0
= L
Y
g.
We investigate
Q
A
(S) =
Z
S
G
A
.
If

Q
A
(S) were to vanish for all vector elds Y , then Q
A
(S) would be a geometric
invariant. This would be so if

G
A
were of the form
G
A
= J1
A
.
with 1
A
a 1-form. For, we would then have
Z
S
G
A
= 0
as dS = 0. However, this is unfortunately false.
Nevertheless, we shall presently show that, in the case that X is a spacetime
translation, if g is asymptotic to

g
at spacelike innity in an appropriate manner, then
G
A
is in fact asymptotically of the formJ1
A
. The case that X is a spacetime rotation
shall be treated later.
Setting
= g
we have
g
= (L
Y
g)
= V
= Y
2
d
2
g
g
2
d
Y
2
g
2
d
Y
2
.
(152)
The last equation holds in any system of coordinates, in particular, in the rectangular
coordinates of

g
.
Nowlet g
= o
2
(r
-
). Here, r =
q
P
3
i=1
(.
i
)
2
andwe saythat a function
of the spacetime coordinates is o
k
(r
-
) if is C
k
and its partial derivatives with
respect to the spacetime coordinates of order l are o(r
--I
), for all l = 0. . . . . k. We
are only allowing gauge transformations which do not affect this fall-off property.
That is, we assume that
= o
3
(r
1-
).
Then we have
g
= d
o
1
(r
-2
). (153)
Note that for translations X
()
=
J
Jx
+
,
G
+2
A
(+)
= J
+2
. (154)
Now, we have
J
+2
2k
d
k
g

o
1
(i
12
)
2k
d
k
g
. (155)
hence
J
+2
2k
d
k
(d
) o
1
(r
-1-2
). (156)
Let T
T
be the largest coordinate ball (r _ 1) contained in S. Then, if _
1
2
,
Z
S
o(r
-1-2
) = o(1
1-2
) 0
as 1 o. It thus sufces to show that there is a 1-form 1
()
such that
1
2
2k
d
k
(d
)

c
2;
= d
;
1
()
d
1
();
. (157)
In fact, dening the totally antisymmetric expressions
1
+;
()
=
1
4
{
(d
;
d
;
(d
;
;
)
;
(d
) ]
we have
2k
d
k
(d
) = d
p
1
+p2
.
Hence, setting
1
()2
=
1
6
1
+;
()
c
;2
.
(157) indeed holds. We conclude the following:
Proposition 4. Let S
i
be the coordinate sphere of radius r on a complete asymptot-
ically at Cauchy hypersurface . Then the limiting quantity
Q
A
(S
o
) := lim
i-o
Q
A
(S
i
) (158)
associated to a translation X is a geometric invariant provided that g
=
o
2
(r
-
) for some _
1
2
.
3.2 Asymptotic atness
What we have just shown implies that the total energy-momentum is a geometric
invariant, provided that the metric g
is asymptotically at in the following sense:

There is a coordinate system in the neighbourhood of spatial innity with respect to
which the metric components satisfy
g
= j
o
2
(r
-
). >
1
2
. (159)
We shall now show that the energy 1 and the linear momentum 1
}
are well dened
under the same assumption. The argument here follows [4]. Since we have shown
these quantities to be independent of the particular choice of such a coordinate system,
we may use coordinates such that the spatial coordinate lines (.
i
= c
i
. i = 1. 2. 3)
are orthogonal to the hypersurfaces .
0
= c
0
, that is we can set
g
0i
= 0.
3.2 Asymptotic atness 55
(We have g
00
=
2
. g
i}
= g
i}
.) We then nd
J
+0i
0
= e
i
. (160)
where
e
i
=
_
det g
4
( g
}n
g
i n
g
i}
g
nn
) d
}
g
nn
. (161)
Thus, the energy is given by
1 = lim
i-o
Z
S
e
i
JS
i
(162)
(in agreement with the ADM expression). Also, we nd
J
+0i
}
=
i
}
=
_
det g
2
( g
in
k
}n

i
}
tr k). (163)
where
i
}
=
_
det g
2
( g
in
k
}n

i
}
tr k) (164)
and k
i}
=
1
2
J
i
Jt
is the 2
nd
fundamental form of the hypersurfaces .
0
= c
0
. Thus,
the linear momentum is given by
1
}
= lim
i-o
Z
S
i
}
JS
i
(165)
(in agreement with the ADM expression).
The hypotheses on , g
i}
and k
i}
corresponding to (159) are
= 1 o
2
(r
-
).
g
i}
=
i}
o
2
(r
-
).
k
i}
= o
1
(r
-1-
). >
1
2
.
(166)
In the following, given a function dened on a hypersurface .
0
= t , we shall
take = o
k
(r
-
) to mean that is a C
k
function of the spatial coordinates and its
partial derivatives with respect to the spatial coordinates of order l are o(r
--I
), for
all l = 0. . . . . k. Moreover, if a compact interval t
1
. t
2
| of values of t is considered,
uniformity of the limit as r o with respect to t is implied. In particular, the
hypotheses (166) are to be meant in this sense.
We rst show that under the hypotheses (166) the limit r oin (162) and (165)
exists. In fact we show a stronger result, namely that there exists a limit independent
of the exhaustion. That is, we do not assume that the exhaustion of is by concentric
coordinate balls. Consider then nested domains T
n
(i.e., T
n1
T
n
) such that
S
n
T
n
= and suppose that the S
n
= dT
n
are C
1
. We claim that
lim
n-o
Z
S
n
e
i
JS
i
= 1. (167)
lim
n-o
Z
S
n
i
}
JS
i
= 1
}
. (168)
To show this recall that for the 4-dimensional spacetime manifold we have
d
=
1
4
1
p
det g 1
+
.
The analogous formula for the 3-dimensional manifold (. g) is
d
i

1
i
=
1
4
1
p
det g

1
+
.
where in fact
1
i
= e
i
.
Also, we have
1
+
=
1
4
g
nn
(
I
}
ni
I
i
n}

I
}
nn
I
i
}i
).
We now appeal to the constraint equation
1 (tr k)
2
[k[
2
= 0
(the twice contracted Gauss equation) to conclude that, under the hypotheses (166),
1 = o
1
(r
-2-2
).
Also, we have
1
+
= o
1
(r
-2-2
).
Consider then two domains T and T
t
with T
t
T such that T contains the coordinate
ball of radius 1. Then we have
Z
S
0
e
i
JS
i

Z
S
e
i
JS
i
=
Z
B
0
\B
d
i
e
i
J
3
.
_ C
Z
B
c
r
-2-2
J
3
.
_ C1
1-2
0 as 1 o.
since >
1
2
. This establishes our claim in the case of 1.
The case of 1
}
is similar. Writing
d
i

i
}
=
1
2
d
i
{
p
det g ( g
in
k
}n

i
}
tr k)].
we see that
d
i

i
}
=
1
2
p
det g {V
i
(k
i
}

i
}
tr k)

I
n
i}
(k
i
n

i
n
tr k)].
We now appeal to the constraint equation
V
i
(k
i
}

i
}
tr k) = 0.
(the contracted Codazzi equation) to conclude that, under the hypotheses (166),
d
i

i
}
=
1
2
p
det g

I
n
i}
(k
i
n

i
n
tr k)
= o(r
-2-2
).
Thus the same argument applies.
We proceed to discuss conservation of the total energy-momentum. Consider rst
the total energy. We have
1(t
2
) 1(t
1
) = lim
i-o
Z
t
2
t
1
Z
S
de
i
dt
JS
i
Jt (169)
and from (161),
de
i
dt
=
dG
i}nn
dt
d
}
g
nn
G
i}nn
d
}
d g
nn
dt
.
where
G
i}nn
=
_
det g
8
( g
in
g
}n
g
i n
g
}n
2 g
i}
g
nn
).
Since
d g
i}
dt
= 2 k
i}
.
we have
2 G
i}nn
d
}
(k
nn
) = 2 G
i}nn
(d
}
k
nn

o(i
2
)
(d
}
)k
nn

o(i
22
)
).
Thus we obtain
de
i
dt
= o(r
-2-
)
hence
r
2
de
i
dt
0
uniformly as r o. Therefore we conclude that
1(t
2
) 1(t
1
) = 0. (170)
Consider next the total linear momentum. We have
1
}
(t
2
) 1
}
(t
1
) = lim
i-o
Z
t
2
t
1
Z
S
d
i
}
dt
JS
i
Jt (171)
and from (164)
d
i
}
dt
=
1
2
p
det g
tr k (k
i
}

i
}
tr k)
dk
i
}
dt

i
}
d tr k
dt
. (172)
(Note that
J
_
det
Jt
= tr k
_
det g.) Appealing to the 2
nd
variational formula
dk
i}
dt
=

V
i
V
}
(
1
i}
2k
in
k
n
}
k
i}
tr k).
and using the assumptions (166), we deduce that
dk
i}
dt
= o(r
-2-
).
Using this we obtain
d
i
}
dt
= o(r
-2-
):
hence
r
2
d
i
}
dt
0
uniformly as r o. Therefore we conclude that
1
}
(t
2
) 1
}
(t
1
) = 0. (173)
3.2.1 The maximal time function. Up to this point we have not made any particular
choice of time function and this arbitrariness is reected in the fact that the lapse
function is not subject to any equation.
We now require the level sets
t
of the time function t to be maximal spacelike
hypersurfaces. That is, any compact perturbation of
t
decreases its volume. Thus
t
satises the maximal hypersurface equation
tr k = 0. (174)
In fact, for a spacetime which is asymptotically at (in the sense above) and satisfying
a certain barrier condition, there exists a unique complete maximal hypersurface
asymptotic to a spacelike coordinate hyperplane at spatial innity. This has been
established by R. Bartnic in [3].
Denition 42. A maximal time function is a time function t whose level sets are
maximal spacelike hypersurfaces which are complete and tend to parallel spacelike
coordinate hyperplanes at spatial innity. Moreover, the associated lapse function
is required to tend to 1 at spatial innity.
Remark. We have one such function (up to an additive constant) for each choice
of family of parallel spacelike hyperplanes in the background Minkowski spacetime.
Two such families are related by the action of an element of the Lorentz group.
We now x the family by requiring
1
i
= 0. (175)
that is, we require the total linear momentum to vanish. Then for any spacetime other
than Minkowski spacetime we obtain a unique time function t (up to an additive
constant) the canonical maximal time function. This is a consequence of the fact that
any non-trivial spacetime has positive energy, which is the positive energy theorem,
to be discussed in the next section. The choice (175) corresponds to the center-of-
mass-frame in Newtonian mechanics.
Let us consider now the Einstein equations relative to a maximal time function,
tr k = 0.
The constraint equations read
Codazzi: V
}
k
i}
= 0. (176)
Gauss:

1 = [k[
2
. (177)
The evolution equations read
1
st
variation:
d g
i}
dt
= 2 k
i}
. (178)
2
nd
variation:
dk
i}
dt
=

V
i
V
}
(
1
i}
2 k
in
k
n
}
) . (179)
Moreover, the trace of the 2
nd
variation equations yields by virtue of the maximality
condition the lapse equation,
. [k[
2
= 0. (180)
Choosing now the canonical maximal time function, we have
1
}
=
1
2
lim
i-o
Z
S
i
}
JS
i
= 0
with
i
}
= k
i
}
. This allows us to impose stronger fall-off on k
i}
. We thus introduce
the notion of a strongly asymptotically at initial data set.
Denition 43. A strongly asymptotically at initial data set is an initial data set
(
M. g. k) such that:
1.

M is Euclidean at innity.
2. There exists a coordinate system in the neighbourhood of innity in

M (that
is on

M \ K) in which the metric components satisfy
g
i}
=
1
2M
r

i}
o
2
(r
-1
). (181)
3. In the same coordinate system we have
k
i}
= o
1
(r
-2
). (182)
The total energy is then given by
1 = 4 M. (183)
3.2.2 Positivity of the energy. We shall now discuss the positive energy theorem of
Schoen and Yau.
Theorem 7 (Positive energy theorem, SchoenYau [26]).
1. Under the assumption
g
i}
=
1
2M
r

i}
O
2
(r
-2
). (184)
it holds that
M _ 0. (185)
2. If also the remainder is O
4
(r
-2
), then
M = 0 (186)
implies that g is the Euclidean metric.
Here and in the following = O
k
(r
-o
) signies that is a C
+
function and its
partial derivatives of order l are O(r
-o-I
) for all l = 0. . . . . k.
Proof. The theoremconcerns strongly asymptotically Euclidean 3-manifolds (
M. g)
with non-negative scalar curvature

1 _ 0. To simplify the notation, we drop the
overlines in the present section.
Proof of Part 1. M _ 0. This consists of three steps.
Step 0: Given such g with M < 0, we can nd another metric g close enough to g
so that also

M < 0 and such that

1 > 0 everywhere.
Step 1: If M < 0, then barriers would exist for minimal surfaces which can be used
to construct a complete area-minimizing minimal surface.
Step 2: Using the 2
nd
variation formula for area we show that the existence of the
minimal surface in Step 1 contradicts the fact that 1 > 0 everywhere.
Step 0. Under a conformal change of g,
g =
4
g. (187)
the scalar curvature

1 of g is related to the scalar curvature 1 of g by
1 =
-5
(1 8.). (188)
Here .is the Laplacian of the metric g. Let us choose a smooth positive function
on

M such that
= O
2
(r
-4
).
We then wish to nd such that
1 =
-4
1 c .
where c is a positive constant. We shall show that this is possible for suitably small c.
Now, by (188), is subject to the equation
. =
1
8
c
5
.
Setting = 1 , this equation reads
. =
1
8
c (1 )
5
.
The implicit function theorem applies if c is suitably small to give us the existence
of a solution such that
= O
2
(c r
-1
).
In particular, = 1 is everywhere positive and
M = M O(c) < 0. (189)

if c is suitably small. In fact, we have

M = M 2N, where
N = lim
i-o
(r) =
1
4
Z
..
Step 1. This is based on the following lemma.
Lemma. Let S be a C
2
minimal hypersurface (that is, a critical point for the area
functional), in an n-dimensional smooth Riemannian manifold (M. g) (dimS =
n 1), which is compact with boundary. Let be a C
2
function on M such that the
values z, for z z
0
. o), are non-critical and the corresponding level sets
2
have
positive mean curvature with respect to the unit normal pointing in the direction of
increase of . Then _ z
0
on dS implies _ z
0
on S.
Proof of the lemma. We consider the restriction

of to the hypersurface S. If the
lemma is not true, then the subset Uof S, where

> z
0
is non-empty, Uis open;
then attains a maximum z

1
> z
0
at a point
U. We have
(
V

)(
) = 0
and
(
V
2

)(
) _ 0.
Choosing a local frame eld (1
o
. a = 1. . . . . n 1) for S in a neighbourhood of
in S, we write
V
o
V
b

for

V
2
(1
o
. 1
b
).
We have
V
o
V
b

= 1
o
(1
b

) (
V
T
u
1
b
)

= 1
o
(1
b

) (V
T
u
1
b
)

where is the orthogonal projection to S,

V
T
u
1
b
= V
T
u
1
b
g(N. V
T
u
1
b
) N.
with N the unit normal to S. Moreover, we have
g(N. V
T
u
1
b
) = g(V
T
u
N. 1
b
)
= 0
ob
with 0 the 2
nd
fundamental form of S. We thus arrive at the formula
V
o
V
b

= V
o
V
b
0
ob
N (190)
on S.
Next we take the trace of (190) with respect to the induced metric g, g
ob
=
g(1
o
. 1
b
), to obtain
.

= g
ob
V
o
V
b
. (191)
recalling that tr 0 = 0 as S is a minimal hypersurface. At an interior maximum point
, the Hessian matrix (
V
o
V
b

)(
) is negative semi-denite. We thus have

(
.

)(
) = ( g
ob
V
o
V
b

)(
) _ 0. (192)
Consider, on the other hand, the level sets
2
of in M. The unit normal vector
eld to
2
, N
t
, pointing in the direction of increase of is
N
ti
= g
i}
d
}
[V [
.
Now, since
is a critical point of

, the tangent planes T
]
^
S andT
]
^
2
1
coincide:
N
t
(
) = N(
).
Therefore, the induced metric is the same at
. Let (1
t
o
. a = 1. . . . . n 1) be a
local frame eld for
2
in a neighbourhood of
coinciding at
with the frame

eld (1
o
. a = 1. . . . . n 1) for S. Then the 2
nd
fundamental form of
2
, 0
t
ob
, is
given in this frame eld by
0
t
ob
= g(V
T
0
u
N
t
. 1
t
b
)
= g
i}
1
tk
o
V
k
g
iI
d
I
[V [
1
t}
b
= 1
tk
o
1
tI
b
V
k
V
I
[V [
1
tk
o
1
tI
b
d
I
d
k
[V [
[V [
2
.
The last term vanishes. We thus obtain
0
t
ob
=
V
2
(1
t
o
. 1
t
b
)
[V [
.
In particular, at
we have 1
t
o
= 1
o
hence
0
t
ob
(
) =
(V
o
V
b
)(
)
[V(
)[
.
Therefore the following holds:
( g
ob
V
o
V
b
)(
) = tr 0
t
(
) [V(
)[ > 0. (193)
In view of equation (191), (193) contradicts (192). This establishes the lemma.
We now consider the circles C
c
with o a constant on the coordinate plane .
3
= 0
in the neighbourhood of innity on the manifold (
M. g).
C
c
= {(.
1
. .
2
. 0) : (.
1
)
2
(.
2
)
2
= o
2
].
We can nd a surface of least area S
c
spanning C
c
. Taking a sequence o
n
] o, and
we want to nd barriers which will allow us to conclude that we can extract a conver-
gent subsequence and thus, passing to the limit, obtain a complete minimal surface S.
C
c
.
3
= 0
.
3
= h
.
3
= h
Figure 11
We apply the lemma to the functions = (.
1
)
2
(.
2
)
2
and = .
3
. In the rst case
the level sets are the coordinate cylinders 1
c
= {(.
1
. .
2
. .
3
) : (.
1
)
2
(.
2
)
2
= o
2
]
and in the second case the level sets are the coordinate planes .
3
= z, the 1
2
. For
1
c
the mean curvature is
1
o
O(r
-2
) > 0
for sufciently large o. For 1
h
and 1
-h
, where h is a positive constant, the mean
curvature is
2Mh
r
3
O(r
-3
). (194)
If M < 0, this is also > 0, provided that h is taken suitably large. In fact, the .
1
and
.
2
are coordinates on the planes 1
2
and we have
V
o
V
b
.
3
= I
3
ob
=
M
r
3
(.
o
b3
.
b
o3
.
3
ob
) O(r
-3
).
Applying then the formula
tr 0
t
= g
ob
V
o
V
b
[V [
yields (194). Let
1
c,h
= {(.
1
. .
2
. .
3
) : (.
1
)
2
(.
2
)
2
_ o and h _ .
3
_ h].
By virtue of the lemma, the surfaces S
c
n
are contained in 1
c
n
,h
. We can then appeal to
interior regularity theory for minimal surfaces to obtain uniforminterior C
3
estimates
and thus conclude that we can extract a subsequence S
c
n
i
converging uniformly on
compact domains to a complete C
2
area minimizing surface S. This completes the
proof of Step 1.
Proof of Step 2. The surface constructed in Step 1 leads to a contradiction when
considering the 2
nd
variational formula for area. A variation of a complete surface S
is a 1-parameter family S
t
such that S
0
= S. The family {S
t
] may not be a foliation.
Nevertheless, the same approach as for a foliation applies if we consider a smooth
mapping
h: (c. c) S M.
The curves
h
]
: (c. c) M
with h
]
(t ) = h(t. ) may not be orthogonal to the surfaces
S
t
= h
t
(S)
with h
t
() = h(t. ). We can nevertheless construct an orthogonal family and thus
redene the homotopy h. The following formulas hold relative to such a normalized
homotopy:
=
Z
S
Jj

.
J
Jt
=
Z
S
tr 0 Jj

.
(195)
This is the rst variation of the area. Here is the lapse function (which measures
the normal separation of the S
t
), dened by
d
dt
= N.
where N is the unit normal to S
t
. In fact the 1
st
variation equations
d g
ob
dt
= 2 0
ob
imply
d Jj

dt
= tr 0 Jj

.
The second variation of the area is then
J
2
Jt
2
=
Z
S
d tr 0
dt
( tr 0)
2
d
dt
tr 0
Jj

. (196)
In our case S is a minimal surface.
Denition 44. A minimal surface S is a surface for which the rst variation of area
vanishes.
This is equivalent to
tr 0 = 0. (197)
Then
J
2
Jt
2
t=0
=
Z
S
d tr 0
dt
Jj

. (198)
For a surface of least area one must have
J
2
Jt
2
t=0
_ 0.
To obtain a suitable expression for
J tr 0
Jt
we consider the 2
nd
variation equations
d0
i}
dt
= 0
in
0
n
}

V
i
V
}
1
i1}1
.
(Here, we have an arbitrary local frame (1
i
) for S, complemented with N.) We have
tr 0 = g
i}
0
i}
.
d tr 0
dt
= g
in
g
}n
d g
nn
dt
0
i}
g
i}
d0
i}
dt
= 2 0
i}
0
i}
g
i}
d0
i}
dt
.
g
i}
1
i1}1
= Ric(N. N) (= 1
i}
N
i
N
}
).
Thus, taking the trace of the 2
nd
variation equations we obtain
d tr 0
dt
=
. ([0[
2
Ric(N. N)).
We now substitute for Ric(N. N) from the (twice contracted) Gauss equation (note
that

1
i}
= 1 g
i}
,

1 = 21, with 1 the Gauss curvature of S)
21 (tr 0)
2
[0[
2
= 1 2 Ric(N. N)
to obtain
d tr 0
dt
=
1
1
2
(tr 0)
2
1
2
[0[
2
1
2
1
.
Substituting in (198) and taking into account the fact that tr 0 = 0 at t = 0 yields
J
2
Jt
2
t=0
=
Z
S

.
2
1
1
2
[0[
2
1
2
1
Jj

.
This must be _ 0 (surface of least area). Integrating by parts in the rst term, this
condition reads
Z
S
1
2
Jj

_
Z
S
1
2
([0[
2
1)
2
Jj

Z
S
[
V [
2
Jj

. (199)
Let Q
p
be the closure of the interior of the coordinate cylinder C
p
,
Q
p
= {(.
1
. .
2
. .
3
) : (.
1
)
2
(.
2
)
2
_ j
2
].
Setting
=
1 on S Q
p
,
log(
,
2
)
log p
on S (Q
p
2 \ Q
p
),
0 on S Q
c
p
2
,
we have
Z
S
[
V [
2
Jj

=
Z
S(Q
,
2
\Q
,
)
[
V [
2
Jj

_ C
Z
p
2
p
"
J
Jr
log(
p
2
i
)
log j
!#
2
r Jr
=
C
(log j)
2
Z
p
2
p
Jr
r
=
C
log j
0 as j o.
Taking then in (199) the limit j oyields
Z
S
1 Jj

_
Z
S
1
2
([0[
2
1) Jj

> 0. (200)
(Recall that by Step 0 we can assume that 1 > 0 everywhere.) It follows that S is
homeomorphic to a disk. (Recall the Cohn-Vossen inequality:
R
S
1 Jj

_ 2 (S),
where (S) is the Euler characteristic of S.) We then consider the disks D
p
= SQ
p
and apply the GaussBonnet theorem with boundary:
Z
T
,
1 Jj

= 2
Z
JT
,
k Js.
where k is the geodesic curvature of dD
p
. The argument proceeds by showing that
limsup
p-o
Z
JT
,
k Js _ 2.
Taking a sequence j
i
o, achieving the limsup then yields
lim
i-o
Z
T
,
i
1 Jj

_ 0
in contradiction with (200). This establishes Part 1.
Proof of Part 2. M = 0 implies that g is at. The proof has two steps.
Step 0: If M = 0 and 1 does not vanish identically, then there exists a metric g
in the conformal class of g such that

M < 0 (N. OMurchadha and J. York
[21]).
By Part 1 we can then conclude that M = 0 implies that 1 vanishes identically.
Proof of Step 0. Setting g =
4
g we solve the equation

1 = 0 for (see (188)):
8 .
1 =

1
5
= 0.
We solve this linear equation under the asymptotic condition 1 at innity.
Since 1 _ 0 the maximum principle applies and we obtain a function which is
everywhere positive (and less than or equal to 1). Since M = 0 the mass

M of g is
contained in the function . In fact we have
M = 2 lim
i-o
r ( 1)
=
1
2
Z
. Jj
=
1
16
Z
1 Jj
< 0.
This proves Step 0.
Step 1: We are nowgiven a metric g with 1 = 0 and M = 0. Consider the following
variation of g (which may not be admissible because it may not satisfy the condition
that the scalar curvature remains _ 0):
g
t
= g t Ric(g).
with t (c. c) where c is to be chosen suitably small. In general,
J 1(g
t
)
Jt
t=0
=

1
is given in terms of g =
d
I
dt
t=0
by
1 = V
i
V
}
g
i}
.tr g g
i}
1
i}
. (201)
(Here, V is the covariant derivative with respect to g
0
= g.) The derivation of this
formula is the following. Since 1 = g
i}
1
i}
we have
1 = g
in
g
}n
g
nn
1
i}

-
i
T
i
g
i}
1
i}
.
Moreover,
1
i}
= V
n
I
n
i}
V
i

I
n
n}
.
where

I
n
i}
is the corresponding variation of the connection, the tensor eld
I
n
i}
=
1
2
g
nn
(V
i
g
}n
V
}
g
i n
V
n
g
i}
).
Substituting in

1
i}
and taking the trace g
i}

1
i}
the result (201) follows.
Consider now the formula for

1 in the case of the variation g
i}
= 1
i}
. Then
tr g = 1 = 0 and V
}
g
i}
= V
}
1
i}
= 0 which is the twice contracted Bianchi
identity in the case 1 = 0. Then (201) reduces to
1 = [ Ric[
2
.
Next, we correct the family g
t
by a suitable conformal change to obtain an admissible
family g
t
. In fact we require

1
t
= 0. Thus with g
t
=
4
t
g
t
we stipulate (see (188))
.
I

t

1
8
1
t

t
= 0.
t
1 at o.
Here the maximum principle does not apply as 1
t
may not be _ 0 but since 1
0
=
1 = 0, if c is suitably small the problem can be solved by appealing to the implicit
function theorem and yields
t
> 0 everywhere. Now, since 1
i}
= O(r
-3
) the
mass of g
t
is that of g, namely 0. Then the mass of g
t
is contained in the function
t
and is given by
M
t
=
1
16
Z
1
t

t
Jj
I
. (202)
Now for each t (c. c), c suitably small, the metric g
t
dened in this way is
admissible, that is, it satises the hypotheses of the theorem. Thus the result of Part 1
applies and we obtain

M
t
_ 0. Moreover, since

M
0
= M = 0, a minimum is
achieved at t = 0. We thus have
J

M
t
Jt
t=0
= 0.
Since 1
0
= 0 and
0
= 1, the formula (202) gives
J

M
t
Jt
t=0
=
1
16
Z
1 Jj

=
1
16
Z
[ Ric[
2
Jj

.
The vanishing of this implies Ric(g) = 0 hence g is at,

M being 3-dimensional.
This establishes Part 2. And with this the whole proof is completed.
3.3 Angular momentum
Let us recall (see (149), (151), (158)) that the asymptotic quantity associated to the
Killing eld X of the background Minkowski metric is given by
Q
A
= lim
i-o
Z
c
G
+0i
A
JS
i
. (203)
The part of G
+0i
A
which involves the rst derivatives of X vanishes for the angular
momentum (but does not vanish for the center of mass integrals (see below)). Taking
X to be the generator of space rotations about the .
o
coordinate axis, we obtain the
angular momentum component
J
o
= lim
i-o
Z
S
c
ob}
.
b
i
}

=:
i
u
JS
i
. (204)
where
i
}
= k
i
}

i
}
tr k, k
i
}
= g
in
k
n}
, and c
obc
is the fully antisymmetric 3-dimen-
sional symbol.
Let us now assume that we have a complete spacelike hypersurface H which is
strongly asymptotically at. More precisely, let us assume that in an admissible chart
in the neighbourhood of innity in H we have
g
i}
=
1
2M
r

i}
O
2
(r
-1-e
). (205)
k
i}
= O
1
(r
-2-e
) (206)
for some c > 0. This implies that 1
}
= 0.
3.3 Angular momentum 71
3.3.1 Independence fromexhaustion. Let U
2
U
1
be domains in H with U
1

T
T
where T
T
is the coordinate ball of radius 1. Then we have
Z
JU
2
i
o
JS
i

Z
JU
1
i
o
JS
i
=
Z
U
2
\U
1
d
i

i
o
J
3
..
Writing
d
i
i
o
= c
oi}

i
}
c
ob}
.
b
d
i
i
}
.
the rst term is
1
2
c
oi}
(
i
}

}
i
)
and the difference
i
}

}
i
reads
i
}

}
i
= ( g
in

in
)k
n}
( g
}n

}n
)k
ni
.
Thus,
1
2
c
oi}
(
i
}

}
i
) = O(r
-3-e
). In regard to the second term we express
d
i
i
}
= V
i
i
}
I
i
in
n
}
I
n
i}

i
n
.
Since by the Codazzi constraint V
i
i
}
= 0, we then obtain d
i
i
}
= O(r
-4-e
). Thus,
the second term is O(r
-3-e
) as well. Therefore,
Z
U
2
\U
1
d
i

i
o
J
3
.
_ C
Z
U
2
\U
1
r
-3-e
J
3
. _ C
t
1
-e
0. (207)
3.3.2 Conservation of angular momentum. Consider the following limit on each
level hypersurface of the canonical maximal time function:
lim
i-o
r
2
d
i
o
(r)
dt
= T
i
o
(). (208)
for each S
2
R
3
(r. are polar coordinates in R
3
). Then we have
dJ
o
dt
=
Z
S
2
T
i
o
()
i
Jj
(209)
where Jj
is the standard measure on S

2
= {[[ = 1] R
3
. Since tr k = 0,
i
o
reduces to
i
o
=
1
2
c
ob}
.
b
p
det g k
i
}
. (210)
Taking account of the fact that
d
_
det g
dt
=
p
det g tr k = 0.
we then have
d
i
o
dt
=
1
2
c
ob}
.
b
p
det g
dk
i
}
dt
. (211)
Substituting from (178), (179) this becomes
d
i
o
dt
=
1
2
c
ob}
.
b
p
det g (
V
i
V
}

1
i
}
). (212)
Now, only the part of

V
i
V
}

1
i
}
which is O(r
-3
) contributes to the limit T
i
o
.
Under the assumptions (205), (206), the lapse equation (180) implies
= 1
N
r
O
2
(r
-1-e
).
where N =
1
4t
R
H
. Jj

=
1
4t
R
H
[k[
2
Jj

. Hence
V
i
=
N
r
2

i
O
1
(r
-2-e
).
and
V
i
V
}
=
(
i
}
3
i
}
) N
r
3
O(r
-3-e
).
Also (205) implies
1
i
}
=
M
r
3
(
i
}
3
i
}
) O(r
-3-e
).
It follows that T
i
o
() is given by
T
i
o
() =
1
2
c
ob}

b
(N M) (
i
}
3
i
}
)
=
1
2
(N M) c
obi

b
.
which implies
T
i
o
()
i
= 0. (213)
In view of the formula (209) the conservation of angular momentum follows.
3.4 The center of mass integrals
The center of mass integrals C
}
correspond to the vector elds
X
(})
= .
}
d
dt
t
d
d.
}
(214)
3.4 The center of mass integrals 73
which generate Lorentz boosts. We have (see (149), (160), (163))
G
+0i
A
= J
+0i
0
.
}
t J
+0i
}
1
+0i
A
(215)
= e
i
.
}
t
i
}

1
4
{ (1
-1
p
det g)
i
}
(
i
}

p
det g g
i}
) ].

=: q
i
The center of mass integrals C

}
are then given by
C
}
= lim
i-o
Z
S
G
+0i
A
JS
i
= lim
i-o
Z
S
(e
i
.
}
t
i
}
q
i
}
) JS
i
. (216)
Remark. Note that
lim
i-o
Z
S
i
}
JS
i
= 1
}
and 1
}
= 0 by (206). Thus, (216) simplies to
C
}
= lim
i-o
Z
S
(e
i
.
}
q
i
}
) JS
i
. (217)
We are goingtopresent a sketchof the proof that the limit (217) exists for anexhaustion
by coordinate spheres.
Sketch of the proof. Given r
2
> r
1
we express
Z
S
2
(e
i
.
}
q
i
}
)JS
i

Z
S
1
(e
i
.
}
q
i
}
)JS
i
=
Z
B
2
\B
1
d
i
(e
i
.
}
q
i
}
)

=:o
J
3
.
=
Z
i
2
i
1
Z
S
2
o
}
(r) Jj
r
2
Jr

=:T(i
1
,i
2
)
.
The hypotheses (205) and (206) imply that
o
}
(r) =
1
}
r
3
O(r
-3-e
)
where 1 is a quadratic expression in M, N, and thus depends only on t . Since
Z
S
2
}
Jj
= 0.
we then obtain
Z
S
2
o
}
(r) Jj
= O(r
-3-e
).
It follows that
[D(r
1
. r
2
)[ _ C
Z
i
2
i
1
r
-1-e
Jr
_ C
t
r
-e
1
0. as r
1
o.
This proves that the limit (217) exists.
We now investigate the gauge invariance of the C
}
; that is, the invariance of the
C
}
under spatial gauge transformations. Recall that thus we are subjecting the metric
g to a diffeomorphism but we are not subjecting the background Minkowski structure
to any transformation. Moreover, we are only allowing transformations which do not
change the asymptotic form (205) and (206).
However, we do not expect the C
}
to be invariant under those transformations
which are asymptotic to non-trivial space translations at innity. For, in Classical
Mechanics the center of mass integrals are affected by space translations. We thus
restrict ourselves to transformations which are asymptotic to the identity at innity.
Such transformations are generated by vector elds Y whose components Y
i
in an
admissible coordinate system in the neighbourhood of innity satisfy
Y
i
= O
3
(r
-e
).
The variations of and g
i}
are
= L
Y
= Y
i
d
i
= O
1
(r
-2-e
). (218)
g
i}
= L
Y
g
i}
= Y
k
d
k
g
i}
g
k}
d
i
Y
k
g
ik
d
}
Y
k
= d
i
Y
}
d
}
Y
i

=O
2
(i
1=
)
O
1
(r
-2-e
). (219)
We then obtain
e
i
=
1
4
(
in
}n

i}
nn
) d
}
(d
n
Y
n
d
n
Y
n
) O(r
-3-e
)
=
1
4
d
k
(d
k
Y
i
d
i
Y
k
) O(r
-3-e
).
(220)
Furthermore we have
q
i
}
=
1
4
{ (
p
det g)

i
}
(
p
det g g
i}
)
] O(r
-2-e
)
and
(
p
det g)
= d
k
Y
k
O(r
-2-e
).
(
p
det g g
i}
)
=
i
}
d
k
Y
k
d
i
Y
}
d
}
Y
i
O(r
-2-e
).
Therefore we obtain
q
i
}
=
1
4
(2
i
}
d
k
Y
k
d
i
Y
}
d
}
Y
i
) O(r
-2-e
). (221)
The terms of order O(r
-3-e
) in (220) and O(r
-2-e
) in (221) do not contribute to

C
}
.
Thus, one nds
C
}
= lim
i-o
1
4
Z
S
{d
k
(d
k
Y
i
d
i
Y
k
) .
}
2
i
}
d
k
Y
k
d
i
Y
}
d
}
Y
i
] JS
i
(222)
= lim
i-o
1
4
Z
B
d
i
{d
k
(d
k
Y
i
d
i
Y
k
).
}
2
i
}
d
k
Y
k
d
i
Y
}
d
}
Y
i
] J
3
..
(223)
Now d
i
d
k
(d
k
Y
i
d
i
Y
k
) = 0 and d
i
.
}
=
i
}
. Therefore the integrant in the volume
integrals is
d
k
(d
k
Y
}
d
}
Y
k
) 2d
}
d
k
Y
k
d
i
d
i
Y
}
d
i
d
}
Y
i
= 0. (224)
which proves the gauge invariance of the C
}
.
Remark. In the expression for C
}
we can replace q
i
}
by q
i
}
:
q
i
}
=
1
4
{(1
p
det g)
i
}
(
i
}

p
det g g
i}
)]. (225)
obtained by replacing by 1 in the expression for q
i
}
.
Proof of the remark. Under a variation of we have
C
}
= lim
i-o
Z
S
q
i
}
JS
i
=
1
4
lim
i-o
Z
S
(
-2
i
}
g
i}
)
p
det g JS
i
.
(226)
Now, with
=
N
r
O(r
-1-e
)
the following hold:
-2
i
}
g
i}
= 2
N
r
i
}

2M
r
i
}
O(r
-1-e
).
=
N
r
O(r
-1-e
).
Consequently,
C
}
=
1
2
lim
i-o
Z
S
=:C

N (N M)
r
2
JS
}
=
C
2
Z
S
2
}
Jj
= 0
which proves the above remark.
We have thus reached the conclusion that the
C
}
= lim
i-o
Z
S
(e
i
.
}
q
i
}
) JS
i
are invariant under gauge transformations which are asymptotic to the identity at
innity. They correspond in Classical Mechanics to the components of the moment
of a mass distribution about a given origin, the product of the total mass times the
components of the position vector of the center of mass relative to that origin.
3.4.1 Conservation of center of mass integrals. Consider
J C
}
J t
=
Z
S
2
D
}
() Jj
. (227)
where
D
}
() = lim
i-o
r
2
de
i
dt
.
}
d q
i
}
dt
(r).
From (225) we obtain
d q
i
}
dt
=
1
4
p
det g
d g
i
}
dt
=
1
2
p
det g k
i}
= O(r
-2-e
).
(Recall that for a maximal time function we have
J
Jt
_
det g = 0.) Similarly,
de
i
dt
=
p
det g
1
4
( g
i n
g
}n
g
i}
g
nn
) d
}
(d
t
g
nn
)

=2k
mn
O(r
-4-e
)
and
d
}
( k
nn
) = V
}
( k
nn
) I
I
}n
k
In
I
I
}n
k
nI
= V
}
( k
nn
) O(r
-4-e
)
= V
}
k
nn
O(r
-4-e
).
Hence we obtain
de
i
dt
=
1
2
p
det g (V
}
k
i}
g
i}
V
}
tr k) O(r
-4-e
)
= O(r
-4-e
).
the rst term vanishing by virtue of the Codazzi equations.
The above imply that
D
}
() = lim
i-o
r
2
de
i
dt
.
}
d q
i
}
dt
(r ) = 0. (228)
We conclude that
d C
}
d t
= 0. (229)
We have thus established the conservation of the center of mass integrals.
4 The global stability of Minkowski spacetime
In this chapter we shall rst state the problem of the global stability of Minkowski
spacetime. We shall then treat simpler analogous problems arising in eld theories in
a given spacetime. Finally we shall conclude the volume with a sketch of the proof
of the global stability theorem of Minkowski spacetime.
4.1 Statement of the problem
The Minkowski spacetime is the simplest solution of the Einstein equations. This is
the spacetime manifold of Special Relativity,
(R
4
. j). where j
= diag(1. 1. 1. 1) in rectangular coordinates.

This is geodesically complete: Every geodesic can be continued ad innitumin afne
parameter.
The function .
0
corresponding to any rectangular coordinate systemis a canonical
maximal time function. The level sets
t
(.
0
= t ) are maximal spacelike hyper-
surfaces. Here they happen to be totally geodesic (not only is tr k = 0 but k
i}
= 0
identically). They are also globally parallel, that is the lapse function is = 1
identically. (Recall the equation

. [k[
2
= 0.)
Cauchy problem with initial data on a complete asymptotically at maximal
hypersurface. Consider initial data sets (H
0
. g
0
. k
0
) with H
0
diffeomorphic to R
3
and tr k
0
= 0, which are strongly asymptotically at. That is, there exists a coordinate
system in the neighbourhood of innity in which the metric coefcients obey
g
0
i
=
1
2M
r
i}
O
2
(r
-1-e
). c > 0. (230)
and we have
k
0
= O
1
(r
-2-e
). (231)
Note that the total linear momentum of such an initial data set vanishes: 1
i
= 0.
The initial data set is, moreover, required to satisfy the constraint equations
Codazzi equations:

V
i
k
i}
= 0. (232)
Gauss equation:

1 = [k[
2
. (233)
4.1 Statement of the problem 79
Our hypotheses are that we are given such an initial data set. Now the problem is
the following:
Problem. Supplement these hypotheses by a suitable smallness condition and show
that we can then construct a geodesically complete solution of the Einstein equations,
tending to the Minkowski spacetime along any geodesic.
4.1.1 Field theories in a given spacetime. Consider Lagrangians of the form
1 = 1
+
Jj
.
where 1
+
is a scalar function, constructed out of the elds, their exterior or more
generally covariant derivatives, the metric and connection coefcients. We give the
following three examples.
Example 1 (Scalar eld ). Let
o = g
.
1
+
= 1
+
(o).
Here only exterior derivatives are involved. So, the Lagrangian does not depend on
the connection coefcients.
Example 2 (Electromagnetic eld J
). Here the spacetime manifold is 4-dimen-

sional and we have
J = J (J
= d
).
= J
: J
= g
k
g
2
J
k2
.
= J
+
J
:
+
J
=
1
2
J
k2
c
k2
.
1
+
= 1
+
(. ).
Also here, only exterior derivatives are involved and the Lagrangian does not depend
on the connection coefcients.
Example 3 (Problem in Riemannian Geometry). Let (M. g) be a compact Riemann-
ian manifold. Find a vector eld U generating an approximate isometry of (M. g).
The solution is the following: Consider
= L
U
g.
80 4 The global stability of Minkowski spacetime
the deformation tensor corresponding to U. This measures the deviation of the 1-pa-
rameter group of diffeomorphisms generated by U from a group of isometries. We
then minimize
Z
[[
2
Jj
under the constraint

Z
[U[
2
Jj
= 1.
The EulerLagrange equation is the eigenvalue problem
div z U = 0.
that is
V
}

i}
z U
i
= 0.
where z is the Lagrange multiplier or eigenvalue. For the analogous problem on a
spacetime manifold
1
+
Jj
Jj
= g
g
2
k2
.
= L
U
g
= V
.
U
= g
.
Here 1
+
depends on the covariant derivatives of U, therefore the connection coef-
cients corresponding to g.
This problem has a direct application to a problem in General Relativity. The
problem in General Relativity is that of preservation of symmetry. Namely, to show
that if the initial conditions possess a continuous isometry group, then the solution
also possesses the same isometry group. The difculty in General Relativity, which
is absent for the analogous problem in the case of a theory in a given spacetime, is to
extend the action of the group fromthe initial hypersurface to the spacetime manifold.
More precisely, the problem can be formulated as follows. Given an initial data set
(H. g. k) for the Einstein equations for which there exists a vector eld

U on H such
that L
U
g = L
U
k = 0, the problem is to extend

U to a vector eld U dened on the
maximal development (M. g) so that U is a Killing eld of g: L
U
g = 0. It turns
out that a suitable way to dene U is to require that it satises the EulerLagrange
equation corresponding to the above Lagrangian, namely the equation
div = 0.
For a eld theory in a given spacetime, the Lagrangian not depending on any other
underlying structure on the manifold other than the metric and the corresponding
connection coefcients, we can dene the gravitational stress T
by considering
the response of the action
AU| =
Z
U
1 Jj
to variations of the underlying metric g. We dene T
by the condition that
A =
1
2
Z
U
T
Jj
for all variations g
with support in U, for any domain Uwith compact closure in

M. By denition T
is symmetric. Thus
T
= T
.
Proposition 5. By virtue of the EulerLagrange equations for the matter elds, T
is conserved. That is
V
= 0.
Proof. Let us denote by the collection of matter elds. Consider any smooth
vector eld X with compact support in U. This generates a 1-parameter group {
t
]
of diffeomorphisms of Uonto itself leaving the action invariant. That is, if we denote
the action by Ag. : U|, then, replacing g by g
t
=
+
t
g and by
t
=
+
t

(schematically; whatever the action of
t
on is; for example, if is a vector eld,
the action is
-t+
, the push-forward by
-t
=
-1
t
) we have
Ag
t
.
t
: U| = Ag. : U|.
So A(t ) := Ag
t
.
t
: U| is independent of t . We thus have
0 =
JA
Jt
t=0
=
Z
U
1
2
T
Jj
= 0.
1 is the variation of A with respect to , which vanishes by virtue of the Euler
Lagrange equations. Moreover, one has
g = L
A
g. g
= V
.
Thus we nd
0 =
Z
U
1
2
T
(V
) Jj
=
Z
U
T
Jj
=
Z
U
(V
) X
Jj
.
As this holds for arbitrary smooth X of compact support in U, the result follows.

Let us write down the gravitational stress corresponding to the three examples
above.
Example 1 (Scalar eld). We have
do
dg
= d
.
Then
T
= 2
J1
+
Jo
d
1
+
g
.
Example 2 (Electromagnetic eld). We have
d
dg
= 2 J
k
J
k
and
d
dg
=
1
2
g
. in view of the identity J

k
+
J
k
=
1
4
g
.
(It is straightforward to check this identity in an arbitrary orthonormal frame.) The
gravitational stress is then given by
T
= 4
d1
+
d
J
k
J
k
d1
+
d
1
+
.
Example 3. In this case,
1
+
=
.
Under a variation of g,
= L
U
g.
We then have
A =
Z
U

2
k

k
1
2
k2
k2
g
Jj
= L
U
g
= U
k
V
k
g
g
k
V
U
k
g
k
V
U
k
.
and, integrating by parts,
Z
U
Jj
=
Z
U
{ V
k
(U
k
)
k
V
k
U
k
V
k
U
] g
Jj
.
We thus nd that
T
= 4
k

k
k2
k2
g
4 { V
k
(U
k
)
k
V
k
U
k
V
k
U
].
Let us now consider the conformal properties.
Proposition 6. If the action in a domain Uis invariant under conformal transforma-
tions of the metric which differ fromthe identity in a subdomain with compact closure
in U, then the gravitational stress is trace-free.
For, with
g = C
2
g.
where C differs from 1 in a domain with compact closure in U, let
A g: U| = Ag: U|.
Then with
g = z g. where z = 2 C

C
has compact support in U, we have
A = 0.
Since
A =
1
2
Z
U
T
Jj
.
this reads
0 =
1
2
Z
U
ztr T Jj
.
As this holds for an arbitrary function z with compact support in U, it follows that
tr T = 0.
Example (Maxwell Lagrangian for electromagnetic theory in a vacuum, 1
+
=
1
4
).
Under the transformation g C
2
g we have
J
.
+
J

+
J
.
1
+
C
-4
1
+
. 1 = 1
+
Jj
1
+
Jj
= 1.
the spacetime manifoldbeing4-dimensional. Thus the actionis conformallyinvariant.
Then
T
= J
k
J
k
1
4
J
k2
J
k2
g
and indeed we have

tr T = 0.
Now let X be a vector eld on M. Consider the vector eld
1
= T
.
We have
V 1 = (V
) X
=
1
2
T
(V
).
by virtue of the symmetry of T
. That is,
V 1 =
1
2
T
.
where = L
A
g. Moreover, if T
is trace-free, we can replace
by its trace-free
part
in the above formula. In the case that dimM = 4 this trace-free part is

1
4
g
tr .
We conclude that in general if X generates a 1-parameter group of isometries of
(M. g) (that is, X is a Killing eld), then 1 is divergence-free:
V 1 = 0.
If A is conformally invariant, the same holds if X generates a 1-parameter group of
conformal isometries of (M. g) (that is, X is a conformal Killing eld). Let us recall
here that a diffeomorphism of M is called an isometry if
+
g = g. It is called a
conformal isometry if there exists a positive function C such that
+
g = C
2
g.
In the case that the canonical stress is related to the gravitational stress by
T
+
= T
2
g
2
. (234)
the conservation of 1 is equivalent to Noethers theorem, since 1 then coincides with
the Noether current.
The divergence theorem gives an integral conservation law corresponding to the
differential law
V 1 = 0.
We consider the dual 3-form
+
1, writing
+
1
;
= 1
c
;
. (235)
Then it holds that
V 1 = 0 J
+
1 = 0. (236)
That is, 1 being divergence-free is equivalent to
+
1 being closed.
Let H
1
and H
2
be homologous hypersurfaces in M. If H
1
and H
2
have boundary,
then dH
1
= dH
2
. We can then apply the divergence theorem to the domain bounded
by H
1
and H
2
to obtain
Z
1
2
+
1 =
Z
1
1
+
1.
We can also apply the theorem to two complete Cauchy hypersurfaces H
1
and H
2
as
in Figure 12.
H
2
H
1
Figure 12
If we can showthat the lateral contribution tends to 0, we again obtain the conservation
law
Z
1
2
+
1 =
Z
1
1
+
1.
Otherwise, we can take the lateral hypersurface to be an incoming null hypersurface
as in Figure 13, in which case its contribution will be non-negative by virtue of the
physical requirement which follows. We then obtain the inequality
Z
1
2
+
1 _
Z
1
1
+
1.
H
2
H
1
Figure 13
The following physical requirement is introduced.
Postulate. The energy-momentumtensor (gravitational stress) satises the following
positivity condition. Let T(. ) be the corresponding quadratic form in the tangent
space at each point. Then we have
T(X. Y ) _ 0 (237)
whenever X. Y are future-directed timelike vectors at a point.
Strong version. We have equality only if the eld is trivial at that point.
Remark. The postulate implies that T(X. Y ) _ 0 whenever X. Y are future-directed
non-spacelike vectors at a point.
If now H is a Cauchy hypersurface with unit future-directed timelike normal N,
then
Z
H
+
1 =
Z
H
1
1
Jj

=
Z
H
T(X. N) Jj

. (238)
where 1
1
is the N-component of 1. That is, complementing N = 1
0
with a frame
(1
1
. 1
2
. 1
3
) for H at a point to a frame for M at that point, we expand:
1 = 1
1
N
3
X
i=1
1
i
1
i
. (239)
Then, since g(N. N) = 1, we have
1
1
= g (N. 1) = T (X. N). (240)
The postulate then implies that the integral (238) is _ 0 whenever X is non-spacelike
future-directed.
We shall nowgive an example of a case where the above theory, which is based on
the gravitational stress, does not apply, nevertheless Noethers theorem still applies.
Example (Electromagnetic theory in a medium). Consider a medium at rest in some
Lorentz frame, the properties of which are invariant under the corresponding time
translations. That is, there is a parallel timelike future-directed unit vector eld u
which is the material 4-velocity, and the material properties are invariant under the
group generated by u. Then there are rectangular coordinates (.
0
. .
1
. .
2
. .
3
) such
that
u =
d
d.
0
(.
0
= t )
and
1 = 1 (.
i
. 1
i
. T
i
: i = 1. 2. 3).
where
1
i
= J
0i
. J
i}
= c
i}k
T
k
(
+
J
0i
= T
i
). As usual, J = J, or, in components, J
= d
. The
1
i
and T
i
are the components of the electric and magnetic eld respectively. The
corresponding displacements are dened by
D
i
=
d1
d1
i
.
H
i
=
d1
dT
i
.
The EulerLagrange equations are Maxwells equations (in the absence of charges
and currents):
V D = 0.
V H
dD
dt
= 0.
We also have the condition JJ = 0 which reads
V T = 0.
V 1
dT
dt
= 0.
The simplest case is that of a homogeneous and isotropic medium.
Homogeneous medium. 1 is invariant under space translations:
d1
d.
i
= 0. (241)
thus
1 = 1(1
i
. T
i
: i = 1. 2. 3). (242)
Isotropic medium. 1 is invariant under space rotations.
If the medium is homogeneous and isotropic, then we have
1 = 1 ( [1[
2
. [T[
2
. 1 T ). (243)
Noting that = J
= 2([1[
2
[T[
2
) and = J
+
J
= 41 T, we
see that even in this case the Lagrangian is more general than that of Example 2. This
is due to the fact that the vector eld u is an additional structure on M, besides the
metric g.
Nevertheless, Noethers theorem still applies to the general medium above. The
invariance under time translations gives rise to the conservation of energy:
E =
Z
I
c J
3
.. (244)
where c = 1 D1 is the energy density and
t
is the hyperplane .
0
= t . In fact,
we have the differential conservation law
dc
dt
V = 0. (245)
where is the energy ux,
= 1 H. (246)
the vector eld
c
d
d.
0
V (247)
being the Noether current corresponding to time translations.
We shall conne ourselves in the following to geometric Lagrangians. Such a
Lagrangian possesses the symmetries of the metric, therefore the pullback by an
isometry of a solution of the EulerLagrange equations corresponding to a given
metric is also a solution of the same equations. In the case that the EulerLagrange
equations are linear, the difference of two solutions is also a solution. It follows that
the Lie derivative of a solution with respect to a vector eld generating a 1-parameter
group of isometries is also, in the linear case, a solution, being the limit of a difference
quotient. Moreover, if the action is conformally invariant, the same applies to the
case of a vector eld generating a 1-parameter group of conformal isometries.
Given a eld , consider the derived elds
n
= L
Y
i
1
. . . L
Y
i
n
(248)
with i
1
. . . . . i
n
= {1. . . . . m], and {Y
1
. . . . . Y
n
] a set of generators of the mparameter
subgroup of the isometry (or conformal isometry) group of g (Killing or conformal
Killing vector elds of g). The construction of the previous paragraph gives a pos-
itive conserved quantity associated to
n
and to each vector eld X generating a
1-parameter subgroup of the isometry (or conformal isometry) group of g which is,
moreover, non-spacelike and future-directed. If we have enough positive conserved
quantities of this type, then Sobolev inequalities imply the uniformdecay of solutions.
In the non-linear case the Lie derivative with respect to a Killing vector eld
(or conformal Killing vector eld) is no longer a solution of the original Euler
Lagrange equations but it is a solution of the equations of variation, namely the
EulerLagrange equations corresponding to the linearized Lagrangian about the given
solution. The underlying structure on which this linearized Lagrangian depends is
not only the metric but also the background solution. Thus the construction which
we have discussed will not yield a conserved quantity unless the background solution
is invariant under the 1-parameter group in question. In general, we will have error
terms which involve the Lie derivative of the background solution with respect to the
correspondingvector eld. The difference of the quantitycorrespondingtoa spacelike
or null hypersurface from the same quantity corresponding to the initial hypersurface
will be the integral of the error terms over the spacetime domain bounded by these
two hypersurfaces.
The aim in the non-linear case is then to achieve closure: obtain enough positive
quantities such that the error integrals can be bounded in terms of the quantities
themselves. Once closure is achieved the global existence theorem for small initial
data will follow by a continuity argument.
We now consider in more detail the conformal group of Minkowski spacetime.
Conformal group of Minkowski spacetime. This group consists of:
1. Spacetime translations. These form an Abelian group. They are generated by
the vector elds (rectangular coordinates)
T
=
d
d.
. j = 0. 1. 2. 3. (249)
of degree 1. T
generates translations along the j-th coordinate axis.

2. Spacetime rotations (Lorentz transformations). These constitute the Lorentz
group SO(3. 1). They are generated by the vector elds
C
= .
d
d.
d
d.
. j. v = 0. 1. 2. 3. j < v. (250)
of degree 0. Here
.
= j
.
C
generates rotations in the j. v-coordinate plane.

3. Scale transformations (. a., a > 0). They are generated by the vector
eld
S = .
d
d.
(251)
of degree 0, which commutes with the spacetime rotations.
4. Inverted spacetime translations. These also form an Abelian group. They are
generated by the vector elds
1
= 2 .
S (.. .) T
. j = 0. 1. 2. 3. (252)
of degree 1. Here
(.. .) = j
.
The 1
are the generators of the 1-parameter groups

11
1. (253)
where 1
is a translation along the j-coordinate axis

1
. = (.
0
. . . . . .
t. . . . . .
3
).
and 1 is the inversion, a discrete conformal isometry of Minkowski spacetime,
dened below.
The inversion 1 is dened by
1 : . . =
.
( .. .)
. ((.. ,) = j
). (254)
Note that
( .. .)(.. .) = 1.
The inverse is
1
-1
: . . =
.
(.. .)
. (255)
Thus
1
-1
= 1 or 1 1 = id.
We have
d.
d .
( .. .)
2 .
( .. .)
2
.
The pullback by 1 of the Minkowski metric j then reads
(1
+
j
) ( .) =
d.
d .
d.
d .
(.)
= C
-2
( .) j
.
where
C( .) = ( .. .).
Thus we have
1
+
j = C
-2
j. (256)
that is, 1 is a conformal isometry of j.
We shall restrict the inversion mapping to 1
(0) in the .-coordinates. (1
(0) is
the chronological future of the origin.) Then (.. .). ( .. .) < 0. So, C > 0.
Proposition7. The inversion maps 1
(0) (in .) to 1
-
(0) (in .). In fact, the inversion
maps light cones onto light cones.
c
.-coordinates
.-coordinates
c
Figure 14
Proof. The light cone with vertex c is given by
(. c. . c) = 0.
Then we have
( . c. . c) = ( .. .) 2 ( .. c) ( c. c)
=
1
(.. .)
2
(.. c)
(.. .)(c. c)
1
(c. c)
=
(. c. . c)
(.. .)(c. c)
= 0.
This proves the proposition.
In particular, future light cones are mapped onto future light cones and past light
cones onto past light cones.
Remark. The entire causal future of a point c 1
(0) is mapped onto a bounded

region in 1
-
(0).
.-coordinates
.-coordinates
c c
I
Figure 15
The innity in 1
(0) (in .) is mapped onto d1

-
(0) (in .). The cone d1
-
(0)\0 is
the future null innity denoted by I
, a concept introduced by R. Penrose (see [24]),

and the origin 0 is the future timelike innity. Any null geodesic in . is mapped onto
a null geodesic in . with a future end-point on I
. Any timelike geodesic in . is

mapped onto a timelike curve in . ending at 0.
.-coordinates
H
1
hyperboloid
plane
c
O
.-coordinates
Figure 16
The inverse image of a spacelike hyperplane 1,
t = k. k a positive constant
(

t = .
0
: r =
q
P
3
i=1
( .
i
)
2
).
is a spacelike hyperboloid H:
t =
1
2k
1
2k
2
r
2
( t = .
0
. r =
q
P
3
i=1
(.
i
)
2
)
This is a hyperboloid through .
0
=
1
k
on the .
0
-axis, which is asymptotic to d1
(c),
where c is the point .
0
=
1
2k
on the .
0
axis. This hyperboloid is intrinsically a space
of constant negative curvature (2k)
2
.
We shall consider examples where the initial data have compact support as well
as an example where the initial data have non-compact support.
.-coordinates
1
T
.-coordinates
H
O
c
O
t
Figure 17
Given initial data of compact support, let

T be the smallest closed ball on the initial
hyperplane containing the support of the data. We choose the .
0
-axis to be the straight
line in Minkowski space orthogonal to the initial hyperplane through the center of

T.
Let O
t
be the point on the .
0
-axis whose future light cone intersects the initial
hyperplane at d
T. We then translate the origin of the (rectangular) spacetime coor-

dinates to a point O on the .
0
-axis which lies properly to the past of the point O
t
and we perform the inversion mapping relative to this new origin. We dene the
positive constant k so that
1
k
is equal to the interval OO
t
and consider the hyperplane
1 :

t = k in the .-space and the corresponding hyperboloid H in the .-space. H
passes through the point O
t
.
We suppose that our system of equations is derived from a Lagrangian and that it
admits a trivial global solution, at which it is regularly hyperbolic, with characteristic
cones coinciding with the light cones of the underlying Minkowski spacetime.
Then, if the initial data is sufciently close to the trivial data, a solution will exist
in the closed spacetime slab (in the original .-space) bounded in the future by the
initial hyperplane and in the past by the parallel hyperplane through O
t
. Moreover,
by the domain of dependence theorem, the solution is trivial outside the union of the
causal future and causal past of

T, in particular in the non-compact portion of H
lying to the future of the initial hyperplane.
We shall investigate below the conditions under which the Lagrangian transforms
under the inversion map into a regular Lagrangian in the .-space. The corresponding
system of EulerLagrange equations will then be a regular system in the .-space
equivalent to the original system.
We can then consider the Cauchy problem in the .-space with initial data on the
hyperplane 1 coming through the inversion mapping from the induced data on H.
This initial data is trivial in a neighborhood of the intersection 1 I
, therefore
extends trivially on 1 outside this intersection. If the initial data on 1 is sufciently
close to trivial, which is the case if we require the original data (in .-space) to be
suitably close to trivial, then the solution of the transformed system in .-space will
exist in the entire closed slab bounded in the past by 1 and in the future by the parallel
hyperplane through the origin. Transforming then the solution back to the original
.-space, we obtain a global solution for the given initial data. In this way a global
existence theorem for small initial data is proven. This approach was rst generally
expounded in [10].
Setting
g = 1
+
j.
the metric g, given by
g = C
2
g. (257)
coincides, according to (256), with the Minkowski metric j:
g = j.
Note that C = 0 on d1
-
(0) = I
L 0 in .-space, which corresponds to innity in

the .-space. In the case of a scalar eld , we also set
= C
-1
. (258)
Then we have
o = g

= C
2
g
(C

) d
(C

)
= C
4
g
(d
log C) (d
log C).
C = 0 C = 0
.-coordinates
Figure 18
Consider the Lagrangian
1 = o Jj
.
The corresponding EulerLagrange equation is the linear wave equation for in the
metric g:
= 0.
Since
Jj
= C
-4
Jj

.
we then have
1 = g
(d
log C) (d
log C) Jj

= ( o 2 g
log C g
2
d
log C d
log C) Jj

where
o = g

.
Now
2 g
log C = d
2
) g
log C
=

V
2
g
log C)

log C.
Consider the vector eld
V
2
g
log C. (259)
Then
Jj

(260)
is a null Lagrangian: we can subtract it from1without affecting the EulerLagrange
equations. After this subtraction the Lagrangian becomes
1 = { o

2
(
log C g
log C d
log C)] Jj

. (261)
Here

log C g
log C d
log C = C

C
-1
= 0. (262)
For,
C
-1
= j
d
2
d .
d .
1
( .. .)
= 0.
Therefore, we obtain
1 = o Jj

.
Thus, the EulerLagrange equation is equivalent to the linear wave equation for

in
the metric g:

= 0.
Example 1 (Non-linear wave equation in Minkowski spacetime). Consider now a
general Lagrangian of the form
1
+
= 1
+
(o) (1 = 1
+
Jj
).
The corresponding EulerLagrange equation is the non-linear wave equation
V
(G(o)d
) = 0 where G =
J1
+
Jo
. (263)
By subtraction of an appropriate constant from1
+
and multiplication by another such
constant we arrive at 1
+
of the form
1
+
(o) = o o
2
J(o).
where J is a smooth function in a neighbourhood of 0. The contribution to 1 of the
non-linear term in 1
+
is
o
2
J(o) Jj
= ( g
(C

) d
(C

))
2
J(C
2
g
(C

) d
(C

)) Jj

=: N Jj

.
The total Lagrangian is thus equivalent to
( o N) Jj

.
where N is regular at C = 0. In fact, we have
N[
D=0
= ( g
C d
C)
2
2
J (0) = 0
because the hypersurface C = 0, i.e. I
is null:
g
Cd
C = 0 along the hypersurface C = 0.

A particular example of the above kind is the following.
Example (Minimal timelike surfaces in 5-dimensional Minkowski spacetime). Con-
sider the Minkowski metric on R
5
:
(J.
0
)
2
3
X
i=1
(J.
i
)
2
(J.
4
)
2
. (264)
Then an arbitrary graph of .
4
over the 4-dimensional Minkowski spacetime is of the
form
.
4
= (.
0
. .
1
. .
2
. .
3
).
The area element of such a graph is
_
1 o J
4
..
and the equation of minimal surfaces is the EulerLagrange equation corresponding
to this Lagrangian.
Example 2 (Non-linear electrodynamics in Minkowski spacetime). Let
= J
. = J
J
+
and
1 = 1
+
(. ) Jj
. (265)
We stipulate that under a conformal transformation g g = C
2
g, the 2-form
J =
X
~
J
J.
. J.
(266)
remains unchanged. Thus, we have
= g
k
g
2
J
k2
J
= C
4
g
k
g
2
J
k2
J

=
= C
4
.
Expressing
J
+
=
1
2
J
k2
c
k2
=
1
2
g
kp
g
2c
J
pc
c
k2
.
and taking into account the fact that c
k2
, the volume form of g, is related to c
k2
,
the volume form of g, by
c
k2
= C
-4
c
k2
.
we see that
J
+
=
1
2
g
kp
g
2c
J
pc
c
k2
= J
+
.
Therefore,
= g
k
g
2
J
k2
J
+
= C
4
g
k
g
2
J
k2
J
+
= C
4

.
Subtracting from 1
+
an appropriate constant we can bring 1
+
to the form
1
+
= G (. ) H (. ).
where G and H are smooth functions in a neighbourhood of (0. 0). Thus,
1 = { G (. ) H (. )]Jj
.
Since Jj
= C
-4
Jj

, we then have
1 = { G (C
4
. C
4

)

H (C
4
. C
4

)] Jj

.
This is regular at C = 0. In fact,
1[
D=0
= { G (0. 0)

H (0. 0)] Jj

.
For a general electromagnetic Lagrangian, the displacement G
is dened by
G
=
d1
+
dJ
. (267)
(Note: In taking the partial derivative with respect to J
, the equality J
= J
is to be taken into account. Thus, lim

t-0
t
-1
{1
+
(J t

J)1
+
(J)] =
1
2
J1
JT
J+
.)
So, for the Lagrangian giving rise to the linear Maxwell equations,
1
+
=
1
4
.
G
= J
.
In general, in terms of a splitting into time and space (as given by the choice of a time
coordinate .
0
):
J
0i
= 1
i
is the electric eld,
J
i}
= c
i}k
T
k
is the magnetic eld
(J
+0i
= T
i
. J
+
i}
= c
i}k
1
k
), and we have
= 2[1[
2
2[T[
2
. = 21 T.
Then with
D
i
=
d1
+
d1
i
. (268)
H
i
=
d1
+
dT
i
. (269)
the displacement reads:
G
0i
= D
i
electric displacement,
G
i}
= c
i}k
H
k
magnetic displacement
(G
+0i
= H
i
. G
+
i}
= c
i}k
D
k
). Also, setting
; = G
. = G
G
+
(270)
we have
; = 2[D[
2
2[H[
2
. = 2D H.
Let us consider the mapping
J G (271)
We have
d
dJ
= 4 J
.
d
dJ
= 4 J
+
.
hence
G
= 4
d1
+
d
J
4
d1
+
d
J
+
. (272)
(Note that J
++
= J
.) It follows that the mapping

(. ) (;. ) (273)
is given by
; =
"
d1
+
d
d1
+
d
2
#
2
d1
+
d
d1
+
d
.
= 2
d1
+
d
d1
+
d

"
d1
+
d
d1
+
d
2
#
.
(274)
The Maxwell equations
J J = 0 (J = J ) (275)
take the form
V T = 0. (276)
V 1
dT
dt
= 0. (277)
The remaining Maxwell equations are the EulerLagrange equations
J G
+
= 0. (278)
These take the form
V D = 0. (279)
V H
dD
dt
= 0. (280)
The initial conditions consist of the specication of T and D at t = 0 subject to the
constraint equations
V T = 0. (281)
V D = 0. (282)
Initial data (T
0
. D
0
) which are C
o
and of compact support can readily be constructed
by taking curls of C
o
vector elds of compact support.
Example 3 (Gauge theory of a complex line bundle over Minkowski spacetime). We
have a complex line bundle over Minkowski spacetime, equipped with a Hermitian
metric. The bundle is topologically trivial, being diffeomorphic to the product R
4
C.
Let us choose an orthonormal basis section o, that is [o[ = 1. Here [ [ is the norm
corresponding to the Hermitian inner product on each complex line bre. As each
bre has only one complex dimension, only the condition [o[ = 1 has to be fullled.
The imaginary line being the Lie algebra of U(1), one has connection coefcients
i
dened by the covariant derivative

D
o = i
o.
The wave function is a section s of the complex line bundle,
s = o.
where is a complex-valued function in Minkowski spacetime. We have
[s[ = [[.
where [(.)[ is the absolute value of the complex number (.). We then nd
D
s = (D
) o.
where
D
= d
.
The Lagrangian of the theory reads
1 = 1
+
Jj
with
1
+
=
1
2
D
z
4
[[
4
1
4
J
. (283)
Here, J = J, that is J
= d
, iJ being the curvature of the bundle.

The 1-form is identied with the electromagnetic potential and the 2-form J with
the electromagnetic eld. The theory describes a charged scalar eld in interaction
with the electromagnetic eld. The EulerLagrange equations are
V
= J
= Im(
).
D
= z [[
2
.
(284)
Here, J
is the electric current density. Under change of basis section

o e
i_
o.
we have
e
-i_
.
.
(285)
The last pair of transformations are the gauge transformations.
Proposition 8. The equivalence class of 1 is conformally invariant.
Proof. The last term in (283) is
1
4
Jj
which we have shown to be conformally invariant, being equal to

1
4
Jj

.
The middle term in (283) is also conformally invariant, being equal to
1
4
[
[
4
Jj

:

= C
-1
.
The rst term is
1
2
oJj
, where now
o = g

= C
2
g
(C
) D
(C
)
= C
4
g
(D
log C) (D
log C)
= C
4
o

V
.
V
= g
[
2
d
log C
(see (262)). Therefore,
1
2
o Jj
is equivalent to
1
2
o Jj

and the proposition is proven.
The initial conditions consist of the specication of , D
0
, 1, T at t = 0.
These are subject to the constraint equations
V T = 0.
V 1 = j.
(286)
where
j = J
0
= Im(
D
0
)
is the electric charge density. Here we can take all the data except 1 to have compact
support. Writing
1 = U V h.
where U is a C
o
compactly supported divergence-free vector eld, V U = 0, the
function h must satisfy .h = j. Therefore h is a harmonic function outside the ball
of support of the rest of the initial data. Outside the future of this ball in Minkowski
spacetime we have
= 0.
T = 0. (287)
1 = V h. independent of t .
For, (287) is a solution of (284). By the domain of dependence theorem it must be
the solution in the domain in question corresponding to the given initial data.
In 3-dimensional Euclidean space (R
3
. e) we consider the inversion map
. .
t
= i
-1
(.) =
.
r
2
.
where r = [.[. With r
t
= [.
t
[, we have r
t
= r
-1
. Then
i
+
e = C
-2
e.
where C = r
t
2
. Thus, setting g = i
+
e, g
t
= e, we have g
t
= C
2
g. Moreover,
setting also h
t
= C
-
1
2
h, we have
.h = 0 .
t
h
t
= 0.
Now our function h is harmonic in R
3
\
T
T
, where

T
T
is the smallest closed ball
containing the support of the data for , D
0
, T, U and thus the support of j. It
follows that h
t
is analytic in T
T
0 , where 1
t
= 1
-1
. We have a convergent Taylor
expansion at the origin, which represents the innity of the original Euclidean space.
This Taylor expansion
h
t
= a b
i
.
t
i
c
i}
.
t
i
.
t
}

(with .
t
h
t
= 0 =tr c
t
= 0. . . . ) corresponds to the multipole expansion of h.
Translating the origin in Minkowski spacetime to a point along the straight line
orthogonal to the initial hyperplane through the center of the ball

T
T
, a point lying
properly to the past of the point the future light cone of which intersects the initial
hyperplane at d
T
T
, we see that the spacetime inversion map 1
-1
in the exterior
of the causal future and past of

T
T
, takes the static solution discussed above to a
solution of the same equations in the image of this domain which admits an analytic
extension through I
. The arguments outlined previously then apply yielding a

global existence and decay theorem for small initial data. The decay of the original
elds follows trivially from the boundedness of the transformed elds.
4.2 Sketch of the proof of the global stability of Minkowski spacetime 103
4.2 Sketch of the proof of the global stability of Minkowski
spacetime
4.2.1 The problem in General Relativity two main difculties
1. The denition of the energy-momentum tensor appropriate to a geometric La-
grangian, namely by considering the variation of the action with respect to the
underlying metric, clearly fails, because this variation vanishes for the gravi-
tational Lagrangian
1 =
1
4
1 Jj
.
This vanishing is the statement of the EulerLagrange equations for gravitation,
namely the Einstein (vacuum) equations.
An alternative approach is to appeal to Noethers theorem after subtracting an
appropriate divergence relative to a background metric, as we have done in
dening the total energy (1), linear momentum (1
i
), angular momentum (J
i
)
and center of mass integrals (C
i
). Among these the energy has been shown
to be positive. But, the energy (a quantity which scales like length) gives us
control on the solutions only after the isoperimetric constant, a dimensionless
quantity, is assumed under control. Therefore, the energy cannot be used, by
itself, to prove regularity.
2. A general spacetime has no symmetries: the conformal isometry group of a
general spacetime is trivial. Therefore we have no conformal Killing vector
elds at our disposal, to use, in conjunction with energy-momentum tensors,
to construct integral conserved quantities.
4.2.2 Resolution of the rst difculty. The idea of how to overcome the rst dif-
culty is based on the following analogy with Maxwells equations of electromagnetic
theory.
Our aim is to derive estimates for the spacetime curvature which will give us the
necessary control on regularity. The idea is to consider the Bianchi identities
V
1
;je
= 0. ;| a cyclic permutation, (288)
as differential equations for the curvature, and the Einstein equations
1
:= g
= 0 (289)
as algebraic conditions on the curvature. Breaking the connection between the metric
and the curvature, we dene a Weyl eld W
;
on a given 4-dimensional spacetime
manifold (M. g
) to be a tensor eld with the same algebraic properties as the Weyl

(or conformal) curvature tensor. Namely:
W
;
= W
;
= W
;
antisymmetry in the rst two as well as in the
last two indices,
W
;j
= 0 cyclic condition,
and the trace condition
g
= 0.
the analogue of the Einstein equations. The rst two properties imply the symmetry
under exchange of the rst and second pair of indices: W
;
= W
;
.
Given a Weyl eld W we can dene a right dual W
+
as well as a left dual
+
W.
The left dual is dened as
+
W
;
=
1
2
c
;
(290)
by freezing the second pair of indices and considering W as a 2-form relative to the
rst pair. The right dual is dened as
W
+
;
=
1
2
W
c
;
(291)
by freezing the rst pair of indices and considering W as a 2-form in the second pair.
However, by virtue of the algebraic properties of W the two duals coincide:
+
W = W
+
. (292)
We shall thus write only
+
W in the following. Moreover,
+
W is also a Weyl eld. In
fact, the cyclic condition for
+
W is equivalent (modulo the other conditions) to the
trace condition for W and vice versa.
AWeyl eld is subject, in the absence of sources, to the vacuumBianchi differential
equations:
V
W
;je
= 0. (293)
We can write these as
D W = 0 (294)
to emphasize the analogy with the exterior derivative. These are the analogues of the
Maxwell equations
JJ = 0.
However, D is not an exterior differential operator, so D
2
= 0. The equation
D
2
W = 0, a differential consequence of the vacuum Bianchi equations, is in fact the
algebraic condition
1
;
+
W
;
1
;
+
W
;
= 0. (295)
Here, 1
;
is the curvature of the underlying metric g
. One can ask: What about

the analogues of the other Maxwell equations
J J
+
= 0 (in the absence of sources)?
The remarkable fact is that the equations
D
+
W = 0 (D W
+
= 0) (296)
are equivalent to the equations DW = 0. In components, the equations D
+
W = 0
read
V
W
;
= 0. (297)
We shall presently dene an energy-momentum tensor analogous to that for
Maxwells equations:
Q
=
1
2
( J
p
J
p
J
+
p
J
+ p
) (298)
(on a 4-dimensional spacetime manifold). This tensor had already been discovered
by L. Bel and I. Robinson in the case W = 1 of a metric g satisfying the Einstein
vacuum equations. We dene
Q
;
=
1
2
( W
p;c
W
p c

+
W
p;c
+
W
p c

). (299)
We call this tensor the BelRobinson tensor. It is a totally symmetric quartic form in
the tangent space at each point (a 4-covariant tensor eld) which is trace-free with
respect to any pair of indices.
Recall that the electromagnetic energy-momentum tensor satises the positivity
condition
Q (X
1
. X
2
) _ 0
for any pair X
1
. X
2
of future-directed timelike vectors at a point with equality if and
only if J vanishes at that point.
Similarly, the BelRobinson tensor has the property
Q (X
1
. X
2
. X
3
. X
4
) _ 0 (300)
for any quadruplet X
1
. X
2
. X
3
. X
4
of vectors at a point all of which are future-directed
timelike, with equality if and only if W vanishes at this point. The above are the
algebraic properties of Q, all of which follow from the algebraic properties of W.
Moreover, if W is a solution of the vacuum Bianchi equations, then Q is divergence-
free:
V
Q
;
= 0. (301)
(This is analogous to the fact that, in electromagnetic theory, JJ = 0 and J
+
J = 0
imply V
= 0 .)
Suppose nowthat we are given three vector elds X, Y , 7 all of which are future-
directed, non-spacelike and generate conformal isometries of (M. g). We denote
by
(U)
=
1
L
U
g (302)
the trace-free part of L
U
g for anyvector eldU. If U generates conformal isometries,
then
(U)
= 0. (303)
We consider the vector eld
1
= Q
;
X
7
;
.
Then, we have
V
= ( V
;

=0
) X
7
;
Q
;
(V
) Y
7
;
Q
;
X
(V
) 7
;
Q
;
X
(V
7
;
).
(304)
We write
Q
;
( V
) Y
7
;
=
1
2
Q
;
( V
) Y
7
;
=
1
2
Q
;
(A)
7
;
=
1
2
Q
;
(A)

7
;
where we have used the symmetric and trace-free nature of Q, and similarly with X,
Y , 7 replaced by Y , 7, X and 7, X, Y respectively. Thus, recalling again the totally
symmetric nature of Q, (304) becomes
V
=
1
2
{ Q
;
(A)

7
;
Q

;
X
(Y )

7
;
Q
;
Y
(7)

;
]
= 0.
(305)
For,
(A)
=
(Y )
=
(7)
= 0, as X. Y. 7 are by assumption conformal Killing
elds. Dening then the corresponding 3-form
+
1
+
1
k2
= 1
c
k2
.
the fact that 1 is divergence-free is equivalent to the fact that
+
1 is closed:
J
+
1 = 0. (306)
We consider the following three cases.
Case 1. We integrate (306) over the domain bounded by the initial Cauchy hypersur-
face H
0
and a Cauchy hypersurface H
t
to the future of H
0
.
H
0
H
t
Figure 19
face H
0
and an outgoing null hypersurface C
u
.
H
0
C
u
Figure 20
face H
0
and an outgoing null hypersurface C
u
capped in the past by a portion of the
(complete) Cauchy hypersurface H
t
.
C
u
H
0
H
t
Figure 21
We then obtain
Z
H
I
+
1 =
Z
H
0
+
1 in Case 1. (307)
Z
C
u
+
1 _
Z
H
0
+
1 in Case 2. (308)
Z
C
c
u
+
1 _
Z
H
0
+
1 in Case 3. (309)
where C
c
u
is C
u
capped by H
t
in the past. Here, all quantities are non-negative.
In particular,
Z
H
I
+
1 =
Z
H
I
1
1
Jj

I
=
Z
H
I
Q (N. X. Y 7) Jj

I
_ 0.
where N is the future-directed unit (timelike) normal to H
t
and Jj

I
the volume
element of the induced metric g
t
on H
t
.
Now, given a Weyl eld W and a vector eld X, the Lie derivative with respect to
X of W, that is L
A
W, is not in general a Weyl eld, because it does not satisfy the
vanishing trace condition. We can however dene a modied Lie derivative

L
A
W
which is a Weyl eld:
L
A
W
;
= L
A
W
;

1
2
(

W
;

W
;

;
W
W
;
) (310)
1
8
tr W
;
.
Here, as in the preceding,
= L
A
g
and
is the trace-free part of
.
The modied Lie derivative commutes with the Hodge dual,
L
A
+
W =
+

L
A
W. (311)
Conformal properties of the Bianchi equations. We consider next the conformal
properties of the Bianchi equations.
Let be a conformal isometry of the underlying manifold (M. g),
+
g = C
2
g. (312)
If W is a Weyl eld satisfying DW = 0 on (M. g), then C
-1
+
W is also a Weyl
eld satisfying the same equation on (M. g). This follows from the fact, readily
established by a straightforward calculation, that if W is a Weyl eld satisfying the
equation DW = 0 on (M. g), then, for any conformal factor C,

W = C
-1
W is also
a Weyl eld satisfying the equation

D

W = 0 on (M. g), where g = C
-2
g.
Remark. Recall that if W is the conformal curvature tensor of (M. g), then the
conformal curvature tensor

W of (M. g) with g = C
-2
g is
W = C
-2
W.
So the transformation W C
-1
W considered above is not related to this.
Suppose now that X is a vector eld generating a 1-parameter group {
t
] of
conformal isometries of (M. g) (a conformal Killing eld). Then if W is a solution
of the Bianchi equations, so is C
-1
t

+
t
W for each t . By the linearity of the Bianchi
equations
J
Jt
C
-1
t

+
t
W
t=0
=

L
A
W (313)
is likewise a solution of the same equations. We see that the term
1
S
tr W in

L
A
W
comes from the conformal weight C
-1
. Thus, if (M. g) possesses a non-trivial
conformal isometry group, we can derive conserved quantities of arbitrary order by
placing in the role of W the iterated (modied) Lie derivatives,
L
A
i
1
. . .

L
A
i
n
W.
an n-th order Weyl eld. Here, i
1
. . . . . i
n
{1. . . . . m], with m being the dimension
of the conformal group of (M. g) and {X
1
. . . . . X
n
] being the generating conformal
Killing elds.
4.2.3 Resolution of the second difculty. We turn to the second difculty, namely
the fact that a general spacetime has only a trivial conformal isometry group.
The crucial observation here is that a spacetime which arises fromarbitrary asymp-
totically at initial data is itself expected to be asymptotically at at spacelike innity
and at future null innity in general, and also, under a suitable smallness restriction
of the initial data, at timelike innity as well.
We thus expect that, under the present circumstances, the spacetime approaches
the Minkowski spacetime as the time tends to innity. Now, the Minkowski spacetime
possesses a large conformal isometry group. We thus expect to be able to dene in the
limit t othe action of a subgroup, at least, of the conformal group of Minkowski
spacetime, as the action of a conformal isometry group in the limit t o.
Then the problem is to extend this action backwards in time up to the initial
hypersurface in such a way as to obtain an action of the said subgroup globally, which
is globally close to being the action of a conformal isometry group, in the sense that
the deformation tensors of the generating vector elds are globally small, and tend
suitably fast to 0 as t o.
It turns out that we can only dene the action of the subgroup of the conformal
group of Minkowski spacetime corresponding to:
1. The time translations.
2. The scale transformations.
3. The inverted time translations.
4. The spatial rotation group O(3).
This is due to the fact that a non-trivial spacetime, corresponding to asymptotically
at initial data, has a non-zero total mass, therefore a non-zero energy-momentum
vector (which can be considered to be a vector at the ideal point at spacelike innity).
Therefore an O(3) subgroup is singled out which leaves this vector invariant.
The group of time translations is the easiest to dene. This corresponds to the
choice of a canonical time function t . This is the canonical maximal time function
relative to which the (spatial) linear momentum 1
i
vanishes. The generating vector
eld T has already been dened. The integral curves of T are the family of timelike
curves orthogonal to the maximal hypersurfaces H
t
and are parametrized by t . The
corresponding group {
r
] is such that
r
is a diffeomorphism of H
t
onto H
tr
.
The rotation group O(3) is to satisfy the condition that it takes any given hyper-
surface H
t
onto itself. To dene the action of O(3) on H
t
we must dene the orbit of
O(3) through a given point . The construction is accomplished with the introduction
of another function u, which is called an optical function as it is a solution of the
eikonal equation
g
u d
u = 0. (314)
This equation expresses the fact that the level sets C
u
of u are null hypersurfaces.
Then the 2-surfaces of intersection,
S
t,u
= H
t
C
u
. (315)
shall be the orbits of the rotation group O(3) on each H
t
. Moreover, the function u
shall also be used to dene the vector elds S and 1 generating the scale transfor-
mations and inverted time translations respectively.
Construction of the optical function u. Thus, the most essential step is the construc-
tion of the appropriate function u. The construction starts by choosing a 2-surface
S
0,0
, diffeomorphic to S
2
, in the initial hypersurface H
0
. We consider dJ
(S
0,0
),
the boundary of the future of S
0,0
, in the spacetime which is assumed to have been
constructed. This has an outer as well as an inner component. The outer component
is generated by the congruence of outgoing null geodesic normals to S
0,0
and the
inner component is generated by the congruence of incoming null geodesic normals
to S
0,0
. We dene the level set C
0
(the 0-level set of u) to be this outer component.
It is an outgoing null hypersurface.
H
0
S
0,0
C
0
S
t,u
= H
t
C
u
H
t
H
t
S
t,0
S
t
,0
Figure 22
Now, there is considerable freedom in the choice of S
0,0
. However, the choice is
subject to the condition that the null geodesic generators of C
0
have no future end-
points.
We must now dene the other level sets C
u
with u = 0. These shall also be
outgoing null hypersurfaces, thus u will, by construction, be a solution of the eikonal
equation (314)
g
u d
u = 0.
H
t
H
0
S
0,0
S
t
,0
inner
C
0
outer
Figure 23
Consider the surfaces (315)
S
t,u
= H
t
C
u
.
We want to impose the following condition: The surfaces S
t
,u
in H
t
must become
equally spaced as t
+
o. That is, u[
H
I
must tend to minus the signed distance

function from S
t
,0
on H
t
as t
+
o.
u < 0
u = 0
u > 0
S
t
,0
Figure 24
The global stability theorem is established by a continuity argument, in the course
of which we are constructing a spacetime slab bounded in the future by the maximal
hypersurface H
t
. The obvious choice of uon H

t
, namely minus the signed distance

function fromS
t
,0
, is inappropriate, because this distance function is only as smooth
as the induced metric g
t
, not one order of differentiability better, which would

be the maximal possible for a function on (H
t
. g
t
). This loss of one order of

differentiability would result in failure of closure of the estimates. The continuity
argument would then fail.
To overcome this difculty, we dene u on H
t
in a different way, namely by

solving a certain equation of motion of surfaces on H
t
, the initial surface being

S
t
,0
. To keep the discussion simple, we neglect the second fundamental formk
t
of
H
t
, and consider the motion of a surface on a (3-dimensional) Riemannian manifold.

Given a function u on such a manifold, a function whose level sets dene locally
a foliation, we have the associated lapse function
a = ( g
i}
d
i
u d
}
u)
-
1
2
. (316)
which measures the normal separation of the leaves. We can think of a as the normal
velocity of a surface, leaf of the foliation. An equation of motion of surfaces is then
a rule which assigns a positive function a to a given surface. Given a surface S
diffeomorphic to S
2
, let be the function
= 1
1
4
(tr 0)
2
. (317)
where 1 is the Gauss curvature of S. (So, = 0 for a round sphere in Euclidean
space.) The rule which denes the equation of motion is
.
log a =

. (318)
where .
is the Laplacian of the induced metric ; on S. In general, for any function

on S we denote by

the mean value of on S. Equation (318) determines a up
to a positive multiplicative constant. The freedom which is left corresponds to the
freedom in relabeling the level sets of u. (We can remove the freedom by requiring
log a = 0.)
To see what this equation of motion accomplishes, consider the trace of the 2
nd
variation equation for a 2-surface on a 3-dimensional Riemannian manifold, once the
Gauss equation has been employed to express

1
33
, e
3
being the unit outward normal
to S, in terms of

1 and 1. The Gauss equation reads
21 (tr 0)
2
[0[
2
= 2
1

1
33
.
Note also that since u is to decrease outwards, we have e
3
=
1
o
J
Ju
. The general
formula we then obtain is the following:
d tr 0
du
= .
a
1
2
a (
1 [0[
2
(tr 0)
2
21). (319)
Now, neglecting the second fundamental formk of H
t
, the Gauss constraint equation

of the imbedding of H
t
in spacetime becomes simply
1 = 0. (320)
It follows that if a is subject to equation (318) above, (319) reduces to the following
propagation equation for tr 0:
1
a
d tr 0
du
=
1
2
[
0[
2
1
2
(tr 0)
2
[V
log a[
2

. (321)
Here,

0 is the trace-free part of 0. All the curvature terms on the right-hand side have
been eliminated. Moreover, by the GaussBonnet theorem
Z
S
,
1 Jj
;
= 4.
hence
=
1
Z
S
,
Jj
;
=
4
1
1
16
Z
S
,
(tr 0)
2
Jj
;
. (322)
where is the area of S
p
. This propagation equation of tr 0 is to be considered in
connection with the Codazzi equations
V
0
B

1
2
V
tr 0 =

1
3
(323)
(complementing e
3
with (e
: = 1. 2) an arbitrary local frame eld for S). These

form an elliptic system on S for

0 given tr 0. Also, in estimating V
log a from the

equation of motion, we appeal to the Gauss equation which expresses 1 as
1 =
1
4
(tr 0)
2
1
2
[
0[
2

1
33
. (324)
Because of the fact that there are no curvature terms on the right-hand side of the
propagation equation (321) and the fact that one order of differentiability is gained
in inverting the 1
st
order elliptic operator in (323), we are able to obtain estimates
for the second fundamental form 0 of the level sets of u which are of one order of
differentiability higher than the estimates for the curvature assumed. Then the level
sets of u and u itself is shown to be three orders of differentiability smoother than the
curvature or one order of differentiability smoother than the metric, as required.
The meaning of the equation of motion of surfaces. Here, we will consider the
Hawking mass m (see [17]) of a surface S diffeomorphic to S
2
in a 3-dimensional
Riemannian manifold of vanishing scalar curvature:

1 = 0. (The energy of S is
4m.) We rst dene the area-radius r of S by
Area (S) = 4r
2
.
Denition 45. The Hawking mass m of a surface S diffeomorphic to S
2
in a 3-
dimensional Riemannian manifold is
m =
r
2
1
1
16
Z
S
(tr 0)
2
Jj
;
. (325)
If T is a small geodesic ball with center at a point , then
lim
B_]
m
Vol(T)
=
1()
16
. (326)
Recall that for the general (non-vacuum) Einstein equations,
1 = 4 T
00
(327)
on a hypersurface with vanishing second fundamental form. We see that the limit at
a point (326) captures only the energy density of matter. There is no gravitational
contribution to the density at a point, the gravitational energy being of non-local
character.
Proposition 9. For suitable families of surfaces S whose interiors exhaust the 3-
manifold we have
m (S) M. (328)
the total mass, provided the 3-manifold is strongly asymptotically Euclidean.
Proof. We are assuming that
g
i}
=
1
M
2j
i}
O
2
(j
-1-e
). c > 0. (329)
where j = [.[. The term O
2
(j
-1-e
) does not contribute to the limit of m. Thus the
proposition follows if we show that the Hawking mass m(S) of the coordinate sphere
S
p
M as j oin the case of the metric
g
i}
=
1
M
2j
i}
. (330)
the Schwarzschild metric. In fact, for any metric of the form
g =
4
[J.[
2
. = ([.[).
changing to polar coordinates, we have
g =
4
( Jj
2
j
2

;
B
(,) J,
J,
B

standard metric on S
2
). = (j).
The arc length s along the rays from the origin is
s =
Z
p
0
2
(j) Jj.
The induced metric on S
p
is ;
B
= j
2
4

;
B
. Then the second fundamental form
0
B
of S
p
is given by
0
B
=
1
2
d;
B
ds
=
1
2
2
d;
B
dj
=
1
1
j
J
Jj
;
B
.
Hence
tr 0 =
2
j
2
1
2j
J
Jj
and
_
det ; = j
2
4
q
det

;
. Thus, we have
Z
S
(tr 0)
2
Jj
;
= 16
1
2j
J
Jj
2
.
Moreover, since 4r
2
= Area(S
p
) =
R
S
,
Jj
;
= 4j
2
4
we have r = j
2
. We
thus obtain
m (S
p
) =
1
2
j
2
"
1
1
2j
J
Jj
2
#
.
In the particular case = 1

2p
we nd
m (S
p
) = M.
Remark. Suitable family of surfaces may be taken to mean the following:
tr 0 =
2
r
O(r
-1-e
). c > 0. r =
r

4
.
and
[
0[ = O(r
-1-e
).
It is these two properties of the surfaces which are required in the above proof.
Denition 46. Given an arbitrary local foliation with lapse function a, the mass
aspect function of each leaf is
j = .
log a 1
1
4
(tr 0)
2
. (331)
By the GaussBonnet theorem,
R
S
1Jj
;
= 4; hence we have
Z
S
j Jj
;
= 4
1
1
16
Z
S
(tr 0)
2
Jj
;
=
8m
r
.
(332)
Thus
j =
2m
r
3
. (333)
Consider now the variation of m as we move through the foliation. We have
Jm
Ju
=
1
2
Jr
Ju
2m
r
r
32
J
Ju
Z
S
(tr 0)
2
Jj
;
. (334)
Now, the general formula (319) reduces in the case

1 = 0 under consideration to
d tr 0
du
= .
a
1
2
a ( [0[
2
(tr 0)
2
2 1).
We express this in terms of j to obtain
1
a
d tr 0
du
=
1
2
(tr 0)
2
1
2
[
0[
2
[V
log a[
2
j. (335)
Using the fact that
dJj
;
du
= a tr 0 Jj
;
.
we then deduce that
J
Ju
Z
(tr 0)
2
Jj
;
=
Z
S
a tr 0 { [
0[
2
2 [V
log a[
2
2 j ] Jj
;
. (336)
On the other hand, since
8r
Jr
Ju
=
J
Ju
=
Z
S
a tr 0 Jj
;
.
we obtain
Jr
Ju
=
r
2
a tr 0. (337)
Now, in the formula for
d
du
R
S
(tr 0)
2
Jj
;
we have the term
2
Z
S
a tr 0 j Jj
;
.
Writing
j = (j j) j
we have
2
Z
S
a tr 0 j Jj
;
= 8r
2
a tr 0 j =
16m
r
a tr 0. (338)
by (333). Therefore, substituting (336) and (337) in (334),
i
32t
(338) cancels the
rst term on the right in (334). We thus remain with
Jm
Ju
=
r
16
Z
S
a tr 0
1
2
[
0[
2
[V
log a[
2
Jj
;
Z
S
(a tr 0 a tr 0) (j j) Jj
;
.
(339)
where we have made use of the fact that (j j) has vanishing mean to rewrite
Z
S
a tr 0 (j j) Jj
;
as
Z
S
(a tr 0 a tr 0) (j j) Jj
;
.
The rst integral in (339) is _ 0, provided that tr 0 _ 0. Thus, under this condition,
the vanishing of the second integral implies that m is a non-increasing function of u.
The second integral vanishes in the following two cases:
Case 1. a tr 0 = a tr 0: This is inverse mean curvature ow.
Case 2. j = j: This is the equation of motion of surfaces, which we discussed
above and which has the smoothing property required in the continuity argument.
Case 1 is a parabolic equation. The problem can be solved in this case in the
outward direction only.
Case 2 may be thought of as an ordinary differential equation in the space of
surfaces. For, the rule assigning the positive function a to a given surface S according
to (318) assigns a function of the same differentiability class as the surface itself. This
is because in inverting the 2
nd
order elliptic operator .
, two orders of differentiability

are gained. The equation of motion of surfaces can be solved in both directions. We
shall discuss general ordinary differential equations in the space of surfaces in a given
3-dimensional Riemannian manifold at the end of the present section.
The problem in Case 2 is actually solved outward globally (that is, for all u _ 0)
and inward up to a surface of area equal to a given fraction of the area of S
0
(= S
t
,0
).
(Note that S
0
is the 0-level set of u.) It turns out that

1
33
.

1
3
.

1
B
have different
decay properties. For xed u and t
+
large they decay like r
-3
. r
-2
. r
-1
, respectively.
The proof of this semiglobal existence theorem for the equation of motion of
surfaces is based on the hypothesis that there exists a background function u
t
with
level sets S
t
u
0
, such that S
t
0
= S
0
, and suitable assumptions hold on the geometric
properties of this background foliation as well as on the components of

1
i}
in the
decomposition with respect to the unit normal and the tangent plane to S
t
u
0
. Once
the surfaces S
t
,u
have been constructed, we dene C
u
for u = 0 to be the inner
component of the causal past of S
t
,u
. In this way the optical function uis constructed
in the spacetime slab bounded by H
t
and H
0
.
Next we dene the vector elds S (scaling) and 1 (inverted time translation). The
vector eld T (time translation) has already been dened by the maximal foliation.
Consider a surface S
t,u
= H
t
C
u
. At each point on S
t,u
we have two null
normals 1 and 1, respectively outgoing and incoming, normalized by the condition
that their components along T are equal to T . The integral curves of 1 are the null
geodesic generators of the C
u
, parametrized by t .
We dene the function
u = u 2 r.
where
r(t. u) =
r
Area(S
t,u
)
4
.
Note that
T =
1
2
( 1 1 ).
Now let us dene S and 1 according to
S =
1
2
( u 1 u 1 ).
1 =
1
2
( u
2
1 u
2
1 ).
To dene the action O(3) in the spacetime slab, we rst consider the nal maximal
hypersurface H
t
. We consider on H
t
the vector eld U:

U
i
= a
2
g
i}
d
}
u.
The integral curves of U are orthogonal to the level sets of u on H
t
, namely the
surfaces S
t
,u
and are parametrized by u. (So, Uu = 1.) Let {
c
] be the 1-parameter
group generated by U. Then
c
restricts to a diffeomorphism of S
t
,u
onto S
t
,uc
.
In particular,
u
: S
t
,0
S
t
,u
is a diffeomorphism. The pullback to S
t
,0
of the induced metric on S
t
,u
rescaled
by r
-2
, namely
+
u
(r
-2
;)
S
I
,u
.
is shown to converge, as u o(that is, at spacelike innity on H
t
), to a metric
;
t
of Gauss curvature equal to 1. Therefore, (S

t
,0
.
;
t
) is isometric to the standard

sphere. The rotation group O(3) then acts as the isometry group of (S
t
,0
.
;
t
), the
sphere at innity.
We then dene the action of O(3) on H
t
by conjugation: Given a point S

t
,u
and an element O O(3), we consider the integral curve
o
c
()
of U through . As o o, this tends to a point q on the sphere at innity. In
other words since (S
t
,0
.
;
t
) is our model for the sphere at innity, we can simply

identify q with the point
-u
() S
t
,0
. Then Oq (S
t
,0
.
;
t
) is well dened.
Finally, the point O is the point
u
(Oq) S
t
,u
. The vector elds C
(o)
with
a = 1. 2. 3 generating this action then satisfy
U. C
(o)
| = 0.
(o)
C.
(b)
C| = c
obc
C
(c)
and are tangential to the surfaces S
t
,u
. The last equations are the commutation
relations of the Lie algebra of O(3).
This action of O(3) on H
t
is then extended to the spacetime slab bounded by

H
t
and H
0
by conjugation with the ow of 1. The integral curves of 1 are the
null geodesic generators of the hypersurfaces C
u
and are parametrized by t . The
1-parameter group of diffeomorphisms generated by 1 maps the surfaces S
t,u
cor-
responding to the same value of u but different values of t onto each other. Given a
point S
t,u
and an element O O(3), we follow the integral curve of 1 through
at parameter value t to the point
+
S
t
,u
at parameter value t
+
. The action of
O(3) on H
t
dened above leads us to the point O

+
S
t
,u
. Finally, O S
t,u
is dened to be the point at parameter value t along the integral curve of 1 through
O
+
at parameter value t
+
.
H
t
S
t,u
S
t
,u
+
O
+
Figure 25
The C
(o)
also satisfy
1. C
(o)
| = 0.
C
(o)
. C
(b)
| = c
obc
C
(c)
and are tangential to the surfaces S
t,u
. Again, the last equations are the commutation
relations of the Lie algebra of O(3).
General equations of motion of surfaces. We now give a general discussion of
ordinarydifferential equations inthe space of surfaces ina givenRiemannianmanifold
(M. g), outlining howa local existence theoremfor such equations is established. Let
A be a rule which assigns to each surface S in M a positive function A(S) on S,
of the same differentiability class as S. Then the problem we are considering is
the following. Given an initial surface S
0
in M, nd a function u dened in a
neighborhood of S
0
in M, such that u = 0 on S
0
and for each level surface S
u
of u,
[Ju[
-1
S
u
= A(S
u
). (340)
Here Ju denotes the differential of u. To solve this problem we rst consider the
following simpler problem. Given an initial surface S
0
in M and a positive function
a dened in a neighborhood of S
0
in M, nd a function u dened in a smaller
neighborhood of S
0
in M, such that u = 0 on S
0
and
[Ju[
-1
= a (341)
in this neighborhood. In other words, nd locally a function u whose 0-level set is
S
0
and whose associated lapse function is the given function a.
Equation (341) is a special case of the stationary HamiltonJacobi equation
H(J) = 1 (342)
for a function on a manifold M, the conguration space. Here H is the Hamilto-
nian,a function on T
+
M, the phase space, and 1 is the energy constant. A particular
class of Hamiltonians are Hamiltonians of the form
H() =
1
2
[[
2
V(q) V T
+
q
M. Vq M. (343)
A Hamiltonian of this form describes the motion of a particle of mass 1 in a potential
V in M. Equation (341) results if we set
= u. V =
1
2
a
-2
. 1 = 0. (344)
We shall presently discuss how solutions to the general stationary HamiltonJacobi
equation (342) are constructed. We consider the canonical equations associated to
the Hamiltonian H. Let (q
1
. . . . . q
n
) be local coordinates on M. Then for q in the
domain of this chart, we can expand T
+
q
M as
=
i
Jq
i
q
.
The coefcients (
1
. . . . .
n
) of the expansion constitute a system of linear coordi-
nates for T
+
q
M. Then (q
1
. . . . . q
n
:
1
. . . . .
n
) are local coordinates on T
+
M and
the Hamiltonian is represented by a function of these. The canonical equations take
in terms of such local coordinates the form
Jq
i
Jt
=
dH
d
i
.
J
i
Jt
=
dH
dq
i
: i = 1. . . . . n. (345)
Denoting by t (q(t ). (t )) a solution of the canonical equations, a variation
through solutions of a given solution, denoted by t ( q(t ). (t )), satises the
equations of variation
J q
i
Jt
=
d
2
H
d
i
dq
}
q
}
d
2
H
d
i
d
}

}
.
J
i
Jt
=
d
2
H
dq
i
dq
}
q
}
d
2
H
dq
i
d
}

}
.
(346)
The canonical form of T
+
M is the 1-form 0 on T
+
M, given in the above local
coordinates by
0 =
i
Jq
i
. (347)
Evaluating 0 on a variation through solutions ( q. ) of a given solution (q. ) we
have
0 ( q. ) =
i
q
i
. (348)
where the right-hand side is a function of t . Abasic proposition of Classical Mechan-
ics, which readily follows from the above equations, is
J
Jt
(0 ( q. )) = J1 ( q. ) (349)
where 1 is the Lagrangian, which in the Hamiltonian picture is a function on T
+
M,
represented by
1 =
i
dH
d
i
H. (350)
Also J1 is the differential of 1, thus
J1 ( q. ) =
d1
dq
i
q
i
d1
d
i

i
. (351)
Given now a closed surface S
0
in M, we construct a solution of (342) vanishing on
S
0
as follows. To each point q
0
S
0
we associate a covector
0
T
+
q
0
M which is
required to vanish on T
q
0
S
0
and satisfy H(q
0
.
0
) = 1. Let t (q(t : q
0
). (t : q
0
))
be the solution of the canonical equations (345) corresponding to the initial conditions
(q
0
.
0
). We then set along each solution trajectory
(q(t : q
0
)) =
Z
t
0
1(q(t
t
: q
0
). (t
t
: q
0
))Jt
t
1t Vq
0
S
0
. (352)
This denes in a neighborhood of S
0
in M, and obviously vanishes on S
0
. We
shall presently show that satises (342). Consider a curve ; : (1. 1) S
0
on S
0
through the point q
0
: ;(0) = q
0
. Let X T
q
0
S
0
be the tangent vector to this curve
at q
0
: X = ;(0). Consider then the 1-parameter family of solutions of the canonical
equations
{t (q(t : ;(s)). (t : ;(s))) : s (1. 1)].
The derivative with respect to s at s = 0 is a variation through solutions t
( q(t : (q
0
. X)). (t : (q
0
. X))) of the solution t (q(t : q
0
). (t : q
0
)). This varia-
tion is the solution of the equations of variation (346) corresponding to the initial
conditions
q(0: (q
0
. X)) = X. (0. (q
0
. X)) =
J
Js
0
(;(s))
x=0
.
Let us take the derivative of both sides of the equation
(q(t : ;(s))) =
Z
t
0
1(q(t
t
: ;(s)). (t
t
: ;(s)))Jt
t
1t (353)
with respect to s at s = 0. We obtain, for the left-hand side,
d
dq
i
q(t;q
0
)
q
i
(t : (q
0
. X)).
and for the right-hand side,
Z
t
0
d1
dq
i
q
i
(t
t
: (q
0
. X))
d1
d
i

i
(t
t
: (q
0
. X))
Jt
t
=
Z
t
0
J
Jt
t
i
(t
t
: q
0
) q
i
(t
t
: (q
0
. X))
Jt
t
=
i
(t : q
0
) q
i
(t : (q
0
. X))
by (349). Here we have taken account of the fact that
i
(0: q
0
) q
i
(0: (q
0
. X)) =
0
X = 0.
It follows that
d
dq
i
q(t;q
0
)

i
(t : q
0
)
q
i
(t : (q
0
. X)) = 0 VX T
q
0
S
0
. (354)
On the other hand, taking the derivative of both sides of (352) with respect to t we
obtain, for the left-hand side,
d
dq
i
q(t;q
0
)
Jq
i
Jt
(t : q
0
).
and for the right-hand side,
1(q(t : q
0
). (t : q
0
)) 1 = (1 H)(q(t : q
0
). (t : q
0
)) =
i
(t : q
0
)
Jq
i
Jt
(t : q
0
)
by (350) and the rst of the canonical equations (345). For, the canonical equations
imply that the Hamiltonian is constant along trajectories, hence
H(q(t : q
0
). (t : q
0
)) = 1. (355)
It follows that

d
dq
i
q(t;q
0
)

i
(t : q
0
)
Jq
i
Jt
(t : q
0
) = 0. (356)
Equations (354) and (356) together imply
d
dq
i
=
i
. (357)
provided that the set of vectors
{ q
i
(t : (q
0
. X)) : X T
q
0
S
0
]
together with the vector
Jq
i
Jt
(t : q
0
)
span T
q(t;q
0
)
M. Moreover, in viewof (355) and (357), is a solution of the stationary
HamiltonJacobi equation (342).
We shall now show how the above is applied to construct a local solution of the
problem associated to (340). We set up an iteration as follows. Given a positive
function a
n
dened in a neighborhood of the surface S
0
in M, we dene the function
u
n
to be the solution of the stationary HamiltonJacobi equation
H
n
(Ju
n
) = 0. (358)
which is negative in the exterior and positive in the interior of S
0
. Here H
n
is the
Hamiltonian
H
n
=
1
2
[[
2
V
n
. V
n
=
1
2
a
-2
n
. (359)
We then dene the new positive function a
n1
by
a
n1
[
S
n
= A(S
n
) (360)
for each level surface S
n
of the function u
n
. The starting point of the iteration is the
function a
0
= 1, in which case u
0
is the signed distance function fromS
0
on M. The
study of the convergence of this iteration then establishes a limit, lim
n-o
u
n
= u,
which is a local solution of the equation of motion of surfaces (340).
4.2.4 The controlling quantity. Having dened the approximate conformal Killing
elds, we consider the 1-form
1 = 1
0
1
1
1
2
. (361)
where
1
0
= Q (1) (.

1. T. T ). (362)
1
1
= Q (

L
O
1) (.

1.

1. T ) Q (

L
T
1) (.

1.

1.

1). (363)
1
2
= Q (

L
2
O
1) (.

1.

1. T ) Q (

L
O

L
T
1) (.

1.

1.

1)
Q (

L
S

L
T
1) (.

1.

1.

1) Q (

L
2
T
1) (.

1.

1.

1).
(364)
Here,

1 = 1 T is everywhere timelike future-directed (for t _ 0). Also, Q(W)
is the BelRobinson tensor associated to the Weyl eld W. Moreover, O stands for
{
(o)
C: a = 1. 2. 3], the generators of the action of O(3).
We dene
1
1
= sup
t
Z
H
I
+
1. (365)
1
2
= sup
u
Z
C
u
+
1. (366)
Then the controlling quantity is
1 = max{1
1
. 1
2
]. (367)
The quantities 1
1
1
2
are dened in the spacetime slab U
t
=
S
t0,t
j
H
t
. So, 1
depends in fact on t
+
.
Remark. 1
1
and 1
2
vanish for solutions which are invariant under rotations and time
translations. They give us effective control on the solutions because of the following
two facts:
1. The only spherically symmetric solution of the vacuum Einstein equations
besides Minkowski spacetime is the Schwarzschild solution.
2. The only static solution of the vacuum Einstein equations besides Minkowski
spacetime is the Schwarzschild solution.
Note that a static spacetime means a spacetime admitting a hypersurface orthogonal
Killing eld which is timelike at innity. Here the Schwarzschild solution is excluded
in view of the fact that the topology of the maximal hypersurfaces is R
3
. To control
the spacetime curvature in terms of the quantity 1, we consider separately the exterior
region E and its complement, the interior region I. The region E is dened by the
inequality
Area(S
t,u
) _ 0 Area(S
t,0
).
where 0 is a constant, 0 < 0 < 1.
Remark. The last termQ(

L
2
T
1)( .

1.

1.

1) in 1
2
is used in controlling the space-
time curvature in the wave zone, that is, in a neighbourhood of C
0
of the form
S
u-c,cj
C
u
for a xed positive constant c.
Remark. The term Q(

L
S

L
T
1)( .

1.

1.

1) in 1
2
is used to control the top order
derivatives of the spacetime curvature in the interior region.
Note that the vector eld S is timelike future-directed in the interior of C
0
, it is null
future-directed on C
0
, and it is uniformly timelike in the region I. We shall presently
outline how control of the spacetime curvature in the interior region is achieved. We
consider the decomposition of a Weyl eld W with respect to a family of spacelike
hypersurfaces with unit future-directed timelike normal e
0
. Complementing e
0
with
an arbitrary local frame eld (e
i
: i = 1. 2. 3) for these hypersurfaces, we dene:
Denition 47. The electric and magnetic parts of a Weyl eld W with respect to a
given family of spacelike hypersurfaces are the symmetric trace-free tensor elds 1
and H on these hypersurfaces given by
1
i}
= W
i0}0
. (368)
H
i}
=
+
W
i0}0
. (369)
Here we consider the electric-magnetic decomposition relative to the canonical
maximal foliation {H
t
]. The Bianchi equations for a Weyl eld W in the presence of
a source J are the equations
D W = J. (370)
The source J is called a Weyl current. Relative to an electric-magnetic decomposition
these equations take the form
div 1 = j
T
. (371)
curl 1

L
e
0
H = o
T
. (372)
div H = j
1
. (373)
curl H

L
e
0
1 = o
1
. (374)
The right-hand sides here contain lower order terms involving the second fundamental
formand lapse function of the foliation as well as the components of the current. Now
if

L
S
W is already controlled, we can decompose

L
e
0
W into a term proportional to
L
S
W which we can place on the right-hand side plus a termof the form

L
A
W, where
X is a vector eld tangential to the leaves of the foliation and satisfying
[X[

_ 0 < 1. (375)
(0 a positive constant) in the interior region 1. Then we obtain a system of the form
div 1 = j
t
T
. (376)
curl 1

L
A
H = o
t
T
. (377)
div H = j
t
1
. (378)
curl H

L
A
1 = o
t
1
. (379)
This system is uniformly elliptic in 1 by virtue of (375), allowing us to obtain interior
estimates for 1 and H.
4.2.5 The continuity argument. Since the vector elds T , S, 1,
(o)
C: a = 1. 2. 3,
are not exact conformal Killing elds, J
+
1 does not vanish. Thus, the integrals
R
H
I
+
1 and
R
C
u
+
1 differ from an integral over H
0
by error integrals which are
spacetime integrals over the part of the spacetime slab bounded by H
t
and H
0
, or
C
u
and H
0
, of expressions which are quadratic in the Weyl elds W and linear in
the deformation tensors of the vector elds. The point is to estimate these error
integrals in terms of the controlling quantity 1 (see Step 2 below).
We introduce a certain set of assumptions on the main geometric properties of the
twofoliations, namelythe {H
t
] and{C
u
]. These are calledthe bootstrapassumptions.
They involve in particular
1. the quantities
sup
S
I,u
(r
2
1). inf
S
I,u
(r
2
1).
where 1 is the Gauss curvature of S
t,u
and 4r
2
= Area(S
t,u
);
2. the quantities
sup
S
I,u
a. inf
S
I,u
a. sup
S
I,u
. inf
S
I,u
.
where a and are the lapse functions of the two foliations
(a
-2
= g
i}
d
i
ud
}
u.
-2
= g
t d
t ).
The isoperimetric constant of each S
t,u
depends on the quantities 1. The Sobolev
inequalities on each H
t
depend on these as well as the rst of the quantities 2. The
Sobolev inequalities of each C
u
depend on the quantities 1 as well as the second of
the quantities 2. The assumption is that the above quantities differ fromtheir standard
values (here: 1) by at most c
0
.
The above are the most important of the geometric quantities, as they control
the Sobolev constants. There is a long list of additional assumptions involving the
remaining geometric quantities, such as sup
S
I,u
(
i tr 0
2
) and inf
S
I,u
(
i tr 0
2
), where 0 is
the second fundamental form of S
t,u
relative to H
t
. All these are also to differ from
their standard values by at most c
0
.
The continuity argument involves the following four steps. There is in addition a
Step 0, which we shall discuss afterwards. We consider the maximal closed spacetime
slab U
t
=
S
t0,t
j
H
t
for which the bootstrap assumptions on the geometric
properties of the two foliations {H
t
] and {C
u
] hold with a constant c
0
.
Step 1 (Estimate of deformation tensors). We show that the bootstrap assumptions
imply that the deformation tensors of the fundamental vector elds T , S, 1,
(o)
C
are bounded in U
t
, in appropriate norms, by another small constant c

1
(depending
continuously on c
0
and tending to 0 as c
0
0).
Step 2 (Error estimates). Using the result of Step 1 we estimate the error integrals as
follows:
[ error integrals[ _ C c
1
1.
This yields the conclusion
1 _ C D C c
1
1.
Here, 1 is the controlling quantity and D is a quantity involving only the initial data.
So, if c
1
is suitably small, we have
1 _ C D.
Step 3. By analyzing the structure equations of the two foliations, we deduce that
the geometric quantities entering the bootstrap assumptions are in fact bounded by
C1. Therefore, by Step 2, under a suitable smallness restriction on D (the size of the
initial data) we can conclude that the said geometric quantities are in fact bounded by
e
0
2
. Thus the inequalities in the bootstrap assumptions are not saturated up to time t
+
.
Step 4. We extend the solution to the slab
S
tt
,t
j
H
t
, for some suitably small
> 0. We rst extend the optical function u
t
, which was dened on the slab U

t
(with nal data on H

t
the solution of the equation of motion of surfaces starting from

S
t
,0
), by extending its level sets as null hypersurfaces, that is, by extending each null
geodesic generator to the parameter interval t
+
. t
+
|. We use this extension, which
we denote by u
t
t
, or rather its restriction to H

t
, in the role of the background

foliation, on the basis of which we construct new nal data on H
t
by solving the
equation of motion of surfaces on H
t
starting from the surface S

t
,0
. With this
nal data we thenconstruct the newoptical functionu
t
. We consider the geometric

quantities associated to the maximal foliation {H
t
] extended to the interval 0. t
+
|
and to the null foliation {C
u
], where u is now u
t
. By continuity, if is chosen
suitably small, these quantities remain _ c
0
, contradicting the maximality of t
+
unless
of course t
+
= o, in which case the theorem is proved.
We shall now discuss Step 0. This concerns the hypothesis on the initial data.
Take a point H
0
and a positive real number z (representing a length). Let J
]
be
the distance function on H
0
from . Setting
D (. z) = sup
H
0
{ z
-2
(J
2
]
z
2
)
3
[ Ric [
2
]
z
-3
Z
H
0
3
X
I=0
(J
2
]
z
2
)
I1
[
V
I
k[
2
Jj

(380)
Z
H
0
1
X
I=0
(J
2
]
z
2
)
I3
[
V
I
T[
2
Jj

.
we dene the invariant
D = inf
]H
0
,2>0
D(. z) (381)
optimizing the choice of and z. This invariant represents the size of the initial data.
In (380), T is the Bach tensor.
Denition 48. On a 3-dimensional Riemannian manifold the tensor eld given by
T
i}
= c
ob
}

V
o
1
ib

1
4
g
ib

1
(382)
is called the Bach tensor.
Remark. We can write
T
i}
= curl

1
i}
=
1
2
(c
ob
i

V
o
1
b}
c
ob
}

V
o
1
bi
).
(383)
where

1
i}
is the trace-free part of the Ricci curvature of the 3-manifold, namely
1
i}
=

1
i}

1
3
g
i}

1.
Thus, the Bach tensor is symmetric and trace-free.
Theorem 8 (Bach [2]). The vanishing of the Bach tensor is necessary and sufcient
for the 3-manifold to be locally conformally at. That is,
T
i}
= 0 g
i}
=
4
e
i}
. (384)
where e
i}
is at, thus locally isometric to the Euclidean metric.
Recall the hypothesis that (H
0
. g) be strongly asymptotically Euclidean:
g
i}
=
1
M
2[.[
i}
o ([.[
-
3
2
)
in an appropriate coordinate systemin a neighbourhood of innity. The principal part
at spatial innity is conformally at, hence has vanishing contribution to the Bach
tensor.
We now state Step 0.
Step 0. On the basis of the hypothesis that
sup
H
0
{ z
-2
(J
2
]
z
2
)
3
[ Ric [
2
]
is suitably small for some H
0
and z > 0, we show that exp
]
, the exponential
mapping with base point is a diffeomorphism of T
]
H
0
onto H
0
. We then have a
foliation {S
p
] of H
0
by the geodesic spheres with center and radius j. We consider
the following Hodge-type elliptic system for the traceless symmetric tensor eld

1
i}
:
div

Ric =
1
6
J

1 =
1
6
J ([k[
2
). (385)
curl

Ric = T. (386)
(tr

Ric = 0). Here, we denote by

J the differential of a function on H
0
.
Remark. The equation (385) is simply the Bianchi identity:
V
}
1
i}

1
2
d
i

1 = 0.
Note that D(. z) gives us control on k and T, thus on the right-hand sides of (385),
(386). The theory of such elliptic systems then gives us estimates for the components
of

1
i}
, hence for the components of

1
i}
, relative to the foliation by the {S
p
]. Denoting
i
o

}
b
1
ob
= a
ob
.
i
o

1
i}
N
}
= b
o
.
1
i}
N
i
N
}
= c.
we in fact obtain
Z
H
0
(J
2
]
z
2
)
2
[ a[
2
Jj

_ C D (. z).
Z
H
0
(J
2
]
z
2
)
2
[b[
2
Jj

_ C D (. z).
Z
H
0
(J
2
]
z
2
)
2
[c c[
2
Jj

_ C D (. z).
where c is the mean value of c on S
p
and a is the trace-free part of a. Moreover, we
have
tr a c =

1 = [k[
2
and
cj
3
2M as j o(j = J
]
[
S
,
).
4.2.6 Estimates for the geometric quantities associated to the maximal foliation
fH
t
g. We shall now sketch how Step 3 is accomplished. We begin with the recovery
of the bootstrap assumptions concerning the maximal foliation {H
t
]. This is done
by considering the structure equations of the foliation.
1. The intrinsic geometry of the H
t
is controlled through the contracted Gauss
equation
1
i}
k
in
k
n
}
= 1
i}
. (387)
Here, 1
i}
is the electric part of the Weyl curvature.
2. The extrinsic geometry of H
t
is controlled through the uncontracted Codazzi
equations
V
i
k
}n

V
}
k
in
= c
n
i}
H
nn
. (388)
Here, H
i}
is the magnetic part of the Weyl curvature.
3. The lapse function of the maximal foliation is controlled through the lapse
equation
. [k[
2
= 0. (389)
The contraction of the Codazzi equations gives the constraint equation
V
}
k
i}
d
i
tr k = 0. (390)
Here, tr k = 0, consequently equations (388) are equivalent to the following Hodge-
type system for the symmetric trace-free tensor eld k:
div k = 0.
curl k = H.
(391)
This is seen from the following remark.
Remark. Let S
i}
be a symmetric 2-covariant tensor eld on a 3-manifold (
M. g
i}
).
Consider
c
ob
i

V
o
S
b}
= C
i}
.
Then the antisymmetric part
C
i}
C
}i
is equivalent to its dual
1
2
(C
i}
C
}i
) c
i}
n
.
This is equal to
c
i}
n
c
ob
i

V
o
S
b}
= ( g
}o
b
n
g
}b
o
n
)

V
o
S
b}
=

V
}
S
n}
d
n
tr S.
On the other hand the symmetric part is
1
2
(C
i}
C
}i
) = (curl S)
i}
by denition.
4.2.7 Estimates for the geometric quantities associated to the null foliation fC
u
g.
We proceed to discuss the recovery of the bootstrap assumptions concerning the null
foliation {C
u
].
S
t,u
H
t
C
u
e
3
e
e
0
e
-
Figure 26
The geometry of a given null hypersurface C
u
is described in terms of its sections
{S
t,u
] by the maximal hypersurfaces H
t
. Let e
3
be the unit outward normal to S
t,u
in H
t
. Set
U = a e
3
. (392)
The vector eld U is characterized by the properties that it is tangential to the H
t
,
orthogonal to the {S
t,u
] foliation of each H
t
and satises
U u = 1.
Let {
c
] be the 1-parameter group of diffeomorphisms generated by U. Then
c
maps S
t,u
onto S
t,uc
. Let e
0
be the unit future-directed timelike normal to H
t
.
Then
T = e
0
(T t = 1).
Let {[
r
] be the 1-parameter group of diffeomorphisms generated by 1. Then [
r
maps S
t,u
onto S
tr,u
. We introduce the normalized null normals e
, e
-
to S
t,u
by
e
= e
0
e
3
.
e
-
= e
0
e
3
.
(393)
Then
1 = e
(1 t = 1).
1 = e
-
(1 t = 1).
The geometry of a given C
u
is described by:
1. The intrinsic geometry of its sections S
t,u
, that is, the induced metric ; and
corresponding Gauss curvature 1.
2. The second fundamental form of S
t,u
in C
u
. This measures the deformation
of S
t,u
under displacement along its outgoing null normal e
, which is intrinsic
to C
u
. Completing e
. e
-
with e
. = 1. 2, an arbitrary local frame for S

t,u
,
is given by
B
= g (V
e
.
e
. e
B
). (394)
The stacking of the C
u
in the foliation {C
u
] is described by:
3. The function a, where
a
-2
= g
i}
d
i
u d
}
u. (395)
That is, a is the lapse function of the foliation of each H
t
by the traces of the
level sets of u. The deformation of S
t,u
under displacement along its incoming null
normal e
-
, which is transversal to C
u
, is measured by , given by
B
= g (V
e
.
e
-
. e
B
). (396)
We have
= 0 j. = 0 j. (397)
where 0 is the second fundamental form of S
t,u
relative to H
t
,
0
B
= g (
V
e
.
e
3
. e
B
). (398)
and j is the restriction of k, the second fundamental formof the maximal hypersurface
H
t
to S
t,u
,
j
B
= g(V
e
.
e
0
. e
B
) = k(e
. e
B
). (399)
The estimate of k
i}
, therefore in particular of j
B
, has been discussed in the preceding
section. The estimate of
B
shall be outlined below. The intrinsic geometry of S
t,u
is controlled by the equation for the Gauss curvature 1:
1
1
4
tr tr
1
2
= j (400)
with j =
1
4
1(e
-
. e
. e
-
. e
). Here and are the trace-free parts of and

respectively.
We now outline the estimate of . Recall that C
u
is dened to be the inner
component of the boundary of the past of the surface S
t
,u
on the nal maximal
hypersurface H
t
. Now, tr satises the following propagation equation along each

generator of C
u
(parametrized by t ):
1
d tr
dt
= v tr
1
2
(tr )
2
[ [
2
. (401)
Here
v =
1
2
g(V
e
C
e
. e
-
). so V
e
C
e
= ve
. (402)
We have
v = V
3
log . = k
33
. (403)
The propagation equation (401) is considered with a nal condition on H
t
, namely
tr for the surface S
t
,u
.
Important fact. By virtue of the Einstein equations no curvature term appears on
the right-hand side of (401).
The propagation equation (401) is considered in conjunction with the null Codazzi
equations:
V
B

B

1
2
V
tr = c
B

B

1
2
c
tr
. (404)
Here c
= k
3
and
=
1
2
1(e
. e
. e
-
. e
). The equations (404) constitute an

elliptic system for , given tr . Because of the fact that there are no curvature terms
on the right-hand side of the propagation equation for tr and the fact that one order
of differentiability is gained in inverting the 1
st
order elliptic operator (div
acting on
trace-free symmetric 2-covariant S
t,u
tensor elds) in (404), we are able to obtain
estimates for which are of one order of differentiability higher than the estimates
for the spacetime curvature assumed.
To estimate a we consider the S
t,u
-tangential 1-form
= V
log a c. (405)
We have
=
1
2
g (V
e
. e
). (406)
We also dene
z
= g (V
e
C
e
-
. e
). (407)
We have
z = V
log c. (408)
Now, with the projection to S
t,u
, it holds that
e
. e
-
| = 2 ( z). (409)
Thus (z) is the obstruction to integrability of the distribution of orthogonal timelike
planes
{ (T
]
S
t,u
) J : M ]. (410)
This manifests itself in the non-commutativity of the 1-parameter groups generated
by 1 and U:
1. U| = a ( z) = 0. (411)
We consider the mass aspect function, dened in general by
j = div
1
1
4
tr tr . (412)
This reduces to the expression (331) when k
i}
= 0. The Hawking mass is dened in
general by
m =
r
2
1
1
16
Z
S
tr tr J
(413)
and we have
j =
2m
r
3
. (414)
The mass aspect function j satises along the generators of C
u
the following prop-
agation equation:
1
dj
dt
jtr = 2 (V

) 2
1
2
tr (div
z [z[
2
1
2
j)
( z) (V
tr c tr )
1
4
tr [ [
2
z.
(415)
Here,
(V

)
B
=
1
2
(V

B
V
;
B
div
)
is the trace-free part of the symmetrized covariant derivative of in S
t,u
.
Important fact. By virtue of the Einstein equations the right-hand side of (415) does
not contain derivatives of the curvature.
The propagation equation (415) with nal data on H
t
, the mass aspect function

of the surface S
t
,u
, is considered in conjunction with the following elliptic equation
for a on S
t,u
, which is simply a re-writing of the denition of j:
.
log a = j div
c 1
1
4
tr tr . (416)
Because of the fact that there are no curvature terms on the right-hand side of the
propagation equation for j and the fact that two orders of differentiability are gained
in inverting the 2
nd
order elliptic operator .
in (416), we are able to obtain estimates

for a which are of two orders of differentiability higher than the estimates for the
spacetime curvature assumed.
The equation of motion of surfaces on the nal maximal hypersurface H
t
, takes
in the general case the same form as in the special case where k
i}
= 0, namely
j = j: (417)
however j is now given by formula (412).
Remarks. j
B
= k
B
is the leading part of k
i}
, namely the part having the slowest
decay. In fact, j can only be bounded by:
[ j[ _ C { c
0
t
-1
t
-
3
2
-
c
2
0
t
-2
]. (418)
The weight functions t
, t
-
, are dened by
t
=
p
1 u
2
. t
-
=
p
1 u
2
. (419)
Note that
tr j = tr k = 0.
For the remaining components of k
i}
we derive the bounds
[c[ _ C { c
0
t
-2
t
-
1
2
-
c
2
0
t
-2
]. (420)
[[ _ C { c
0
t
-
5
2
c
2
0
t
-2
]. (421)
On the other hand, we derive for the bounds
[ [ _ C c
0
r
-2
.
r
2
tr 1
_ C c
0
r
-1
.
(422)
Consider the behavior on a given C
u
as t o. Then
r j tends to a non-trivial limit.
while
r 0.
Since we have
= 0 j.
= 0 j.
it follows that
r
2
tr 0 1.
while
r

0 r j tends to a non-trivial limit.
r 2 r j tends to a non-trivial limit.
In particular, [
0[
2
,(tr 0)
2
tends to a non-trivial limit. Consequently, the surfaces S
t,u
on a given C
u
do not become umbilical as t o. In fact, the last limit is proportional
to the amount of energy radiated per unit time per unit solid angle at a given retarded
time and a given direction.
4.2.8 Decomposition of a Weyl eld with respect to the surfaces S
t;u
. In con-
cluding our sketch of the proof of the global stability of Minkowski spacetime, we
shall show in detail some of the more delicate estimates in Step 2. The discussion
shall make use of the decomposition of a Weyl eld and its associated BelRobinson
tensor with respect to the surfaces S
t,u
.
Consider the null frame e
. e
-
supplemented by e
. = 1. 2, a local frame eld

for S
t,u
. The components of a Weyl eld W in such a frame are
B
= W (e
. e
. e
B
. e
).
B
= W(e
. e
-
. e
. e
-
).
=
1
2
W(e
. e
. e
-
. e
).
=
1
2
W (e
. e
-
. e
-
. e
).
j =
1
4
W (e
-
. e
. e
-
. e
). oc(e
. e
B
) =
1
2
W (e
. e
B
. e
-
. e
).
. are symmetric trace-free 2-covariant tensor elds on S
t,u
, . are 1-forms on
S
t,u
, j. o are functions on S
t,u
.
Here, c is the area 2-form of S
t,u
.
Each of , has two algebraically independent components,
each of , has two components,
and j, o are two functions.
So, there are ten component-functions in all.
The components of Q(W), the BelRobinson tensor associated to W, in the
(timelike) plane spanned by e
. e
-
, are:
Q (W) (e
-
. e
-
. e
-
. e
-
) = 2 [[
2
. (423)
Q (W) (e
. e
-
. e
-
. e
-
) = 4 [[
2
. (424)
Q (W) (e
. e
. e
-
. e
-
) = 4 (j
2
o
2
). (425)
Q (W) (e
. e
. e
. e
-
) = 4 [[
2
. (426)
Q (W) (e
. e
. e
. e
) = 2 [[
2
. (427)
Note that e
0
=
1
2
(e
e
-
) is the unit future-directed normal to H
t
, and that
T = e
0
=
1
2
(e
e
-
). (428)
1 = 1 T =
1
2
(t
2
t
2
-
e
-
). (429)
the weight functions t
, t
-
being given by (419).
Recall now the controlling quantity 1 = max{1
1
. 1
2
], where 1
1
is dened in
(365) as the supremum over t of an integral on H
t
and 1
2
in (366) as the supremum
over u of an integral on C
u
.
In what follows we shall use the following notation. Let . g be positive functions.
Then ~ g denotes that there exists a constant C > 0 such that:
C
-1
_ g _ C .
Let us consider the integrants in 1
1
and the ones in 1
2
. The integrants in 1
1
are
Q (W)(

1.

1. T. e
0
) for W =

L
O
1.

L
2
O
1.
Q (W)(

1.

1.

1. e
0
) for W =

L
T
1.

L
O

L
T
1.

L
S

L
T
1
and we have
Q(W)(

1.

1. T. e
0
) ~ t
4
-
[[
2
t
2
-
t
2
[[
2
t
4
(j
2
o
2
[[
2
[[
2
) (430)
Q(W)(

1.

1.

1. e
0
) ~ t
6
-
[[
2
t
4
-
t
2
[[
2
t
2
-
t
4
(j
2
o
2
) t
6
([[
2
[[
2
).
(431)
The integrants in 1
2
are
Q (W) (

1.

1. T. e
) for W =

L
O
1.

L
2
O
1.
Q (W) (

1.

1.

1. e
) for W =

L
T
1.

L
O

L
T
1.

L
S

L
T
1
and we have
Q(W)(

1.

1. T. e
) ~ t
4
-
[[
2
t
2
-
t
2
(j
2
o
2
) t
4
([[
2
[[
2
). (432)
Q(W)(

1.

1.

1. e
) ~ t
6
-
[[
2
t
4
-
t
2
(j
2
o
2
) t
2
-
t
4
[[
2
t
6
[[
2
. (433)
4.2.9 The borderline error integrals. We nowshowhowsome of the error integrals
in Step 2, requiring more careful treatment, are handled. These are the following:
1.
Z
U
I
Q (W =

L
O
1 or

L
2
O
1)
;
(

1)

1
;
T
Jj
.
2.
Z
U
I
Q (W =

L
O
1 or

L
2
O
1)
;
(T)

1
;
Jj
.
3.
Z
U
I
Q (W =

L
T
1 or

L
O

L
T
1 or

L
S

L
T
1)
;
(

1)

1
;
Jj
.
For the error integrals 1 the worst term is
Z
U
I
Q (W)
B;
(

1)

B
1
;
T
Jj
because
(

1)

B
contains t
2

B
which is the part with the slowest decay. In fact
this is merely bounded pointwise by Cc
1
. Moreover, the leading part of the integrant
is obtained by taking the part of

1 with the largest weight, namely
1
2
t
2
, and the
1
2
e
-
part of T :
Z
U
I
[Q
(

1)
B-

B
[ t
2
Jj
_ C c
1
Z
U
I
[Q
B-
[ t
2
Jj
.
Now, we have
Q
B-
(W) = { 2

2 ; (j
2
o
2
) ]
B
.
Here, we denote
(.

,)
B
= .
,
B
,
.
B
;
B
(. ,).
We estimate
Z
U
I
[

[ t
2
Jj
_ C
Z
Ju
(
t
-2
-
Z
C
u
t
4
-
[[
2
1
2
Z
C
u
t
4
[[
2
1
2
)
_ C 1
Z
t
-2
-
Ju = C 1.
comparing with (432) and noting that
R
o
-o
t
-2
-
Ju = . Since the geometric mean
of the weights of . in (432) is the weight of (j. o), the term in (j. o) in Q
B-
can be estimated in the same way. We thus obtain
[ error integrals 1 [ _ C c
1
1.
Z
U
I
Q (W)
B;
(T)

B
1
;
Jj
because
(T)

B
contains j
B
which decays pointwise only like t
-1
t
-1
-
. The lead-
ing part of this comes from the part of

1
;

1
of the largest weight, namely from

(
1
2
)
2
t
4
e
;
:
Z
U
I
[Q
B
(T)

B
[ t
4
Jj
_ C c
1
Z
U
I
[Q
B
[ t
3
t
-1
-
Jj
.
Now, we have
Q
B
= { 2 ; [[
2
2 j 2 o
+
]
B
where
+
denotes duality on S
t,u
. We estimate
Z
U
I
[(j. o) [ t
3
t
-1
-
Jj
_
Z
Ju
(
t
-2
-
Z
C
u
t
2
-
t
2
(j
2
o
2
)
1
2
Z
C
u
t
4
[[
2
1
2
)
_ C 1
Z
t
-2
-
Ju _ C 1.
comparing with (432). We thus nd
1
1.
Z
U
I
Q
B;
(

1)

B
1
;
Jj
and the leading part of this comes from (

1
2
)
2
t
4
e
;
, the part of

1
;

1
of the
largest weight. Also,
(

1)

B
is pointwise bounded by Cc
1
. Hence, we have
Z
U
I
[Q
B
(

1)

B
[ t
4
Jj
_ C c
1
Z
U
I
[Q
B
[ t
4
Jj
.
Here, the terms 2;[[
2
and 2(jo
+
) in Q
B
(see formula above) are on equal
footing. We estimate
Z
U
I
[(j. o) [ t
4
Jj
_
Z
Ju
(
t
-2
-
Z
C
u
t
4
-
t
2
(j
2
o
2
)
1
2
Z
C
u
t
6
[[
2
1
2
)
_ C 1
Z
t
-2
-
Ju _ C 1.
comparing with (433). A similar estimate holds for
Z
U
I
[[
2
t
4
Jj
.
We thus nd
1
1.
Note that in the error integrals 1 the principal part is Q
B-
multiplied by t
2

B
.
Thus only the trace-free relative to S
t,u
part of Q
B-
enters. In any case, the
corresponding spacetime trace is 0. That is,
;
B
Q
B-

trace relative to S
I,u
Q
--

=4(p
2
c
2
)
= 0.
The absence of an uncontrollable term linear in each of , , is an instance of
the following general identity: For any three symmetric trace-free 2-dimensional
matrices , T, C we have
tr( T C) = 0.
Equivalently, there is no product in the space of symmetric trace-free 2-dimensional
matrices, because for any two such matrices , T we have
T T tr( T) 1 = 0.
The leading role played by symmetric trace-free 2-dimensional matrices can be traced
back to the symbol of the Einstein equations. For, as we have seen in Chapter 2, the
space of dynamical degrees of freedom of the gravitational eld at a point can be
identied with the space of such matrices.
Bibliography
[1] R. Arnowitt, S. Deser, C. W. Misner, Energy and the criteria for radiation in general
relativity. Phys. Rev. 118 (1960), 11001104. 35
[2] R. Bach, Zur Weylschen Relativittstheorie. Math. Z. 9 (1921), 110135. 129
[3] R. Bartnik, Existence of maximal surfaces in asymptotically at space-times. Comm. Math.
Phys. 94 (1984), 155175. 59
[4] R. Bartnik, The mass of an asymptotically at manifold. Comm. Pure Appl. Math. 39
(1986), 661693. 54
[5] L. Bel, Introduction dun tenseur du quatrime ordre. C.R. Acad. Sci. Paris 248 (1959),
12971300.
[6] H. Bondi, M. G. J. van de Burg, A. W. K. Metzner, Gravitational waves in general relativity.
Proc. Roy. Soc. London Ser. A 269 (1962), 2152.
[7] J. Cheeger, D. Ebin, Comparison theorems in Riemannian geometry. North-Holland, New
York 1975.
[8] Y. Choquet-Bruhat, Thorme dexistence pour certain systmes dequations aux drives
partielles nonlinaires. Acta Math. 88 (1952), 141225. 19
[9] Y. Choquet-Bruhat, R. P. Geroch, Global aspects of the Cauchy problem in general rela-
tivity. Comm. Math. Phys. 14 (1969), 329335. 25
[10] D. Christodoulou, Global solutions for nonlinear hyperbolic equations for small data.
Comm. Pure Appl. Math. 39 (1986), 267282. 93
[11] D. Christodoulou, Nonlinear nature of gravitation and gravitational wave experiments.
Phys. Rev. Lett. 67, (1991), 14861489. viii
[12] D. Christodoulou, The action principle and partial differential equations. Ann. Math.
Studies 146, Princeton University Press, Princeton, NJ, 2000. ix, 17, 26, 41
[13] D. Christodoulou, The stability of Minkowski space-time. In Surveys in differential geom-
etry: essays on Einstein manifolds, Surv. Differ. Geom. VI, International Press, Boston,
MA, 1999, 365385.
[14] D. Christodoulou, S. Klainermann, The global nonlinear stability of the Minkowski space.
Princeton Math. Ser. 41, Princeton University Press, Princeton, NJ, 1993. ix
[15] R. Courant, D. Hilbert, Methods of mathematical physics. Vol. 2: Partial Differential Equa-
tions, Wiley Classics Library, John Wiley & Sons, NewYork, 1989.
[16] L. C. Evans, Partial differential equations. Grad. Stud. Math. 19, Amer. Math. Soc., Prov-
idence, RI, 1998.
[17] S. W. Hawking, Gravitational radiation in an expanding universe. J. Math. Phys. 9 (1968),
598604. 114
[18] S. W. Hawking, G. F. R. Ellis, The large scale structure of space-time. Cambridge Monogr.
Math. Phys. 1, Cambridge University Press, London 1973.
144 Bibliography
[19] J. Jost, Riemannian geometry and geometric analysis. Universitext, second edition,
Springer-Verlag, Berlin 1998.
[20] C. W. Misner, K. S. Thorne, J. A. Wheeler, Gravitation. W. H. Freeman and Company, San
Francisco, CA, 1973.
[21] N. OMurchadha, J. W. York, Gravitational energy. Phys. Rev. D 10 (1974), 23452357.
68
[22] E. Noether, Invariante Variationsprobleme. Nachr. d. Knig. Gesellsch. d. Wiss. zu Gttin-
gen, Math-phys. Klasse (1918), 235257. 36
[23] R. Osserman, The isoperimetric inequality. Bull. Amer. Math. Soc. 84 (1978), 11821238.
[24] R. Penrose, Conformal treatment of innity. In Relativit, Groupes et Topologie, 1963
Summer School Theor. Phys. Univ. Grenoble, Les Houches Lectures, Gordon and Breach,
NewYork 1964, 565584. 91
[25] R. Penrose, Techniques of Differential Topology in Relativity. Regional Conf. Ser. Appl.
Math. 7, SIAM, Philadelphia, PA, 1972.
[26] R. Schoen, S. T. Yau, On the proof of the positive mass conjecture in general relativity.
Comm. Math. Phys. 65 (1979), 4576. 60
[27] R. Schoen, S. T. Yau, Lectures on Differential Geometry. Conf. Proc. Lecture Notes Geom.
Topology I, International Press, Cambridge, MA, 1994.
[28] R. M. Wald, General Relativity. University of Chicago Press, Chicago, IL, 1984.
[29] H. Weyl, Raum-Zeit-Materie. Fourth edition, Springer-Verlag, Berlin 1921. 38
[30] E. Witten, A new proof of the positive energy theorem. Comm. Math. Phys. 80 (1981),
381402.
Index
ADM, 35, 55
angular momentum, 36, 70
conservation of, 71
Anti-de-Sitter-space, 3
asymptotically
Euclidean Riemannian manifold,
35
at initial data set, 35
Bach tensor, 129
background metric, 48
BelRobinson tensor, 105
Bianchi equations, 104
conformal properties of, 108
with a current, 126
Bianchi identity, 9, 103
twice contracted, 9
borderline error integrals, 139
boundary current, 50
canonical equations, 121
canonical form, 122
canonical momentum, 36
canonical stress, 36, 84
Cauchy hypersurface, 6, 86
Cauchy problem, 21
with initial data on a complete
maximal hypersurface, 78
causal curve, 2
causal future, 4
causal past, 4
center of mass integrals, 70, 72
conservation of, 76
characteristic
subset, 12
Choquet-Bruhat, 19
Codazzi equations, 32, 78, 113, 131
trace of the Codazzi equations, 32
null Codazzi equation, 134
conservation law, 37, 85
conserved quantities, 38
constraint equation, 32, 59, 78
continuity argument, 127
controlling quantity, 125
current, 41
compatible, 41
curvature tensor, 9, 32
de-Sitter-space, 3
deformation tensors
estimate of, 127
displacement, 97
electric, 98
magnetic, 98
domain of dependence, 24
theorem, 30
eikonal equation, 110
Einstein, 37
tensor, 9
Einstein equations, 9, 32, 103
relative to a maximal time function,
59
Einstein pseudo-tensor, 38
Einstein vacuum equations, 9, 32
EinsteinWeyl energy-momentum
conservation laws, 49
electric eld, 98
elliptic
regular ellipticity, 17
energy, 54
positive energy theorem, 60
total, 35
energy-momentum tensor, 9, 85
equations of variation, 11
error estimates, 128
146 Index
Euclidean at innity, 35
EulerLagrange equations, 10, 46, 81,
94, 99
evolution equations, 59
extrinsic geometry, 131
eld
electric, 98
magnetic, 98
gauge invariance, 39, 74
gauge transformations, 52
Gauss curvature, 133
Gauss equations, 32, 78, 113, 114
once contracted, 131
double trace of, 32
GaussBonnet theorem, 113
General Relativity, 9
geodesics
null, 4
globally hyperbolic, 6
gravitational stress, 81, 84, 85
HamiltonJacobi equation, 121
Hamiltonian, 121
Hawking mass, 114, 135
Hilberts action principle, 37
HopfRinow theorem, 3
hyperbolic
globally, 6
regularly, 17, 25
hyperboloid, 92
intrinsic geometry, 131
inverse mean curvature ow, 118
inverted time translations, 110
Killing eld, 50
Lagrangian, 10, 36, 42, 79, 94, 122
lapse equation, 59, 131
lapse function, 112, 133
of the maximal foliation, 131
level set, 110
Lie ow, 41, 44
light cone, 91
linear momentum, 36, 54
Lorentzian metric, 1
magnetic eld, 98
mass aspect function, 116, 135, 136
maximal
development, 25
spacelike hypersurface, 58
time function, 59
canonical, 59
hypersurface, nal, 119
Maxwell equations, 13, 99
Minkowski spacetime, 48, 78, 109
conformal group of, 89
electromagnetic Lagrangian in, 97
global stability of, 78
sketch of the proof of, 103
non-linear electrodynamics in, 96
motion of matter, 9
equations of, 9
motion of surfaces
equation of, 112, 118
equation of, 136
Noether
domain Noether current, 42
Noether current, 37, 46, 84
conserved, 49
Noether transform, 17
Noethers theorem, 36, 84
for sections of C, 47
in the domain case, 42
null cone, 2
null foliation, 132
null hypersurface, 4, 110
null innity, 91
optical function, 110, 128
Index 147
Poincar group, 49
propagation equation, 134, 136
for tr 0, 113
regularity, 103
regularly hyperbolic, 17
Ricci tensor, 9
scalar curvature, 9
scale transformations, 110
Schwarzschild metric, 115
Schwarzschild solution, 125
sections of tensor bundles over M, 42
space rotations, 36
space translations, 36
spacelike, 2
spacetime, 1
curvature, 103
manifold, 1, 103
spatial rotation group, 110
strongly asymptotically Euclidean, 129
stronglyasymptoticallyat initial data set,
60, 70, 78
structure equations, 131
symbol, 11
time orientability, 2
time translations, 36, 110
inverted, 110
timelike, 2
variation equations
rst variation equations, 31
second variation equations, 31
Weyl, 37
current, 126
eld, 103, 126
decomposition of, 137
electric part of, 126
magnetic part of, 126
curvature tensor, 103

Christodoulou D. - Mathematical - Problems - of

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Christodoulou D. - Mathematical - Problems - of

Uploaded by

Copyright:

Available Formats

SS

2008 European Mathematical Society

(), is the set of all points q M

(), we dene the temporal distance of q from by

(1) and the causal past J

() for some 1].

u = 0. These hypersurfaces are generated by null

(1) have past

(1) may have future end-points, even when 1 is a

u (components with respect to an arbitrary frame), we have

t = 1. Thus the orthogonal curves are parametrized by t . That is, the 1-

the Einstein tensor, 1

the Ricci tensor and 1 the

. From the original Bianchi identity

= 0. The equations are then equivalent to

and the curvature and Ricci tensor components in

are the components of a covector and g

the components of a possible

at a point M and a covector

(.. u(.). du(.))

(.. u(.). du(.)).

, j = 1. . . . . n are local coordinates on an

) u and the characteristic is

is non-trivial for every covector . This degeneracy is

is the electromagnetic potential, a 1-form. The Maxwell

is non-trivial for all

dened on Q. We distinguish two cases: in the rst case

in each equivalence class in Q such that

consists of the spacelike 2-dimensional plane

reads, as noted above,

are the components of an arbitrary covector. Consider then the equiv-

has only trivial null space on Q.

g = 0 and tr g = 0 implies that g lies in the null space of o

can be identied with the space of

are the components of X and the

are the components of . The space

on the velocity varia-

. That is, if X () and z > 0, then zX lies in ().

= 0 is the wave equation.

, j = 0. 1. 2. 3, each of which is a solution of the

= 0 in an arbitrary system of local coordinates

for = 0. 1. 2. 3. we have a solution of the above equation. Since

the components of the Ricci curvature tensor, let us dene

vanish identically provided that the initial conditions vanish, that is

= 0, such that the conditions

of the Cauchy problem for the reduced equations

M, which is to be identied with the initial hypersurface

vanish throughout D(C). Therefore the solution of the

M. g) not assumed) satisfying the constraint equations, gives rise

1. Thus the vacuum Einstein equation

(with j = 1. . . . . n) the independent variables, q

be a vector eld generating a 1-parameter group

the volume form of g. This

are the unknown functions and 1 depends on their second derivatives.

to be the canonical momentum, where

(.)). The canonical stress is then

are constants. Let

are arbitrary constants, the following differential conservation laws hold:

an einer Stelle stets

as energy components of the gravitational eld; for, these quantities

can be made to vanish at

= 0, above) are without a physical

. We say that an integrated quantity Q is gauge invariant if for